International Conference on Innovative Computing and Communications

Context: Virtualization technology has facilitated the entire cloud computing scenario it has manifested the computing environment in varied ways. It enables to create small instances which are provisioned for execution of user application. These small virtual instances are employed for perceiving the potential throughput. However, there arises a need for efficaciously placing the virtual machines providing services, thereby increasing the resource utilization. This placement of virtual machines to the tangible devices or the physical machine is known as Virtual Machine Placement Problem (VMPP). In VMPP numerous VMs are consolidated on fewer physical machines so as to have energy efficient computing. Problem: The problem is to design an optimized technique for resolving the Virtual Machine Placement Problem and achieve reduction in the power consumption and number of VM migrations without violating the SLA. Objective and Focus: To achieve and propose an effective solution for dynamic Virtual Machine Placement Problem considering initial allocation and reallocation. The primary focus is by employing meta-heuristic algorithm, and thereby obtaining the robust solution and achieving the associated QoS parameters. Method: The proposed method is inspired by the peculiar behaviour of cuckoos, where cuckoos search optimal nest for laying its eggs. An algorithm integrated with machine learning has been devised and evaluated with other meta-heuristic algorithms. Result: The proposed optimization algorithm effectively optimizes virtual machine placement and migrations. For instances of 50 to 500 virtual machines, the performance has been evaluated, where for 500 virtual machines, the power consumption is 55.0660916 kW, SLA-V is 0.00208696, and the number of migrations is 36. Conclusion: To sum up, the proposed optimization algorithm, when compared with two competent and recent meta-heuristic algorithms, exhibits exceptional performance in terms of power consumption, SLA-V, and several migrations. Thus, evaluation proves the strength of the proposed algorithm.

People have converged on a worldwide level because of advancements in communication technologies. They are critical in ensuring freedom of speech by allowing individuals to openly express their thoughts, behaviors, and opinions. Although this presents an excellent chance for racism, trolling, and exposure to a flood of nasty online content. As a result, social media’s exponential rise in hate speech is profoundly affecting society. In this study, we used machine learning and deep learning algorithms to identify hate speech, and we compared their performance to create an ensemble model. Researchers gathered and integrated two distinct datasets of hateful tweets in Tamil that were produced by Bharathi Raja Chakravarthi et al. Tweets in this dataset fall into two categories: non offensive and offensive. This dataset contains 10,129 tweets. In addition, the researchers used six selected machine and deep learning algorithms for this study namely. Support Vector Machine (SVM), Logistic Regression (LR), Naïve Bayes (NB), Bidirectional LSTM, Multi-layer Perceptron (MLP) and Multilingual BERT. When it comes to detecting hate speech, SVM (82%) and LR (82%) have the best Accuracy. Furthermore, researchers developed two ensemble algorithms to construct a most efficient model. First ensemble model was created by combining SVM, LR and NB and the second ensemble was developed using SVM and LR. Four algorithms including the two ensemble models were obtained same accuracy. Therefore, the researchers compared the F1 score and found that the ensemble model 02 outperformed other classifiers. The results of this study are significant because they can be used as a benchmark for future research on Tamil language hate speech using various machine learning algorithms to detect hate speech more effectively and accurately.

Chaotic systems have established an important place in the generation of random sequences for the protection of vital confidential data. In the paper, we propose a novel secure hybrid chaotic system ‘MARCS’ consisting of control parameters, modulo operations, and three one-dimensional chaotic maps logistic, sine, and exponential maps combined in a nonlinear manner to achieve high complexity and characteristics necessarily required for high information security applications. We analyze its performance by examining chaotic and randomness properties to evaluate its security strength. The simulation results reveal that the proposed hybrid chaotic system MARCS possesses strong chaotic and randomness properties. Thus, MARCS can be incorporated in the design of secure information systems for meeting the high information security requirements.

Cardiovascular health is becoming more common over the world. Electrocardiography (ECG) is a test that utilizes an electrical signal in the heart to diagnose and monitor cardiac problems. Utilizing the myDAQ data acquisition device and LabVIEW, identifying cardiac illness in real-time is now not only possible but also simple. As a result, this article presents a system that can capture ECG signals in real-time, detect heart problems, and reveal whether a waveform is in range or not utilizing light-emitting diodes (LEDs). LabVIEW software was utilized to create and evaluate the ECG signal of a patient and to calculate associated parameters for the prototype home ECG system described in this article. These parameters include–the Rate of Heart, intervals-of-PR, intervals-of-QT, intervals-of-ST, waves-of-P, and waves-of-T durations. The distant station receives the recorded ECG signal through the Internet, as well as its related factors. Tests were done on the system utilizing ECG’s simulator, web-based-ECG, and actual ECG acquisition.

Due to the advent of smart cities, vehicular technologies are high focus to perform intelligent transmission. Road traffic control and accident control becomes a primary tasks in Vehicular Adhoc Networks (VANETs). In VANETs communication is ineffective without presence of Road Side Units (RSUs). In some places due to lack of RSUs data transmission is improve as well as end to end delay and routing overhead is increased during the time of communication so effective RSU deployment is very essential to provide effective communication in multi-hop VANETs. In this paper, Centralized RSU Deployment Strategy is proposed to control and monitor the traffic density of the vehicle. The parameters which are considered for the process of centralized RSU deployment strategy are centralization scale, past transmission rate and location identification. To analysis the performance of the network the simulation is carried out in NS2.35 and SUMO. The parameters which are taken for the performance analysis are packet delivery ratio, end to end delay, routing overhead and packet loss. In order to execute comparative analysis the recent research works are considered which are AVRCV and RDSTDV. From the outcome it is understood that the proposed CRDSMV approach achieves 70–200 ms lower end to end delay, 7–14% improved packet delivery ratio, and 50–300 packets lower routing overhead and 50–240 Kbps improved throughputs when compared with the earlier methods AVRCV and RDSTDV.

Vehicular ad hoc networks (VANETs) are the newer technology which is used for maximum of the intelligent transmission system (ITS)-based application. Due to the high-speed dynamic nature of VANETs, link failure and data loss can occur during communication between vehicles. Several routing protocols are proposed in VANETs but still constructing the effective routing protocol is still an open research area. In this paper, hybrid routing protocol is recommended, namely cross-layer in connectivity-aware greedy routing protocol (CLCAGRP). This protocol is subdivided into three categories. They are greedy-based routing protocol, connectivity-aware greedy routing protocol (CAGRP), and cross-layer in connectivity-aware greedy routing protocol (CLCAGRP). These methods are mainly used to improve network stability by reducing the connectivity failures and packet loss during transmission. Cross-layer approach is used to find the optimal path between the sources to the destination at the time of data transfer. In order to evaluate the performance of the proposed CLCAGRP, the parameters which are calculated in the simulation evaluation are packet delivery ratio, energy efficiency, network throughput, packet loss, and routing overhead. The results of the proposed CLCAGRP are compared with the earlier research works such as E-GRP and ACO-GRP. The outcome proves that the proposed CLCAGRP achieves (16%) better packet delivery ratio, (15%) better energy efficiency, (600 Kbps) better throughput, (330packets) lower packet loss, (1100 packets) lower overhead when compared with the earlier works.

The amount of textual data has increased significantly over time, creating the potential for machine learning and natural language processing research. Nowadays, sentiment analysis of YouTube comments is a fascinating topic. Due to the inconsistent and poor quality of the data, not much has been done to pre-process the numerous user comments. In this study, we use machine learning approaches to do sentiment analysis on YouTube comments relating to hot subjects. Although several techniques for sentiment analysis have been created recently for the health sector, the field of reverse diabetes has not yet been much investigated based on YouTube video comment analysis. To categorise the dataset, the machine learning techniques like Naive Bayes, support vector machine, logistic regression, and random forest were utilised.

Due to continuous innovations in machinery, digital images are extensively used in numerous domains like remote sensing, military, communication, medical, etc. These images may encompass subtle and private data. Thus, images are mandatory to be secured from illegal access. Several image security methodologies have been developed in previous years. Most of the techniques to secure the images use encryption. In this method, an encrypted image is generated from the plane image using an encryption algorithm and a secret key. Researchers are continuously devising new techniques in various domains for image encryption. To increase image security, various researchers have devised new techniques for image encryption. In this perspective, an analytical survey of the current image encryption approaches is presented in this paper. The investigation covers the application of encryption techniques in various application domains such as finance, defense, remote sensing, medical, etc. The image encryption techniques reviewed in this paper are categorized based on chaotic map, DNA coding, neural network, and spatial techniques. These techniques are compared to assess those on the basis of various security parameters. In addition, the performance of these techniques is also analyzed. The future directions to encourage researchers for devising new encryption techniques are also presented in this paper.

It is impossible for a prosperous IT sector to avoid underestimating the amount of work, the amount of money, and the amount of time their projects would take. The introduction of Agile, (PBPM)principle-based process models, such as Scrum, brought about a substantial shift in the industry. This shift in culture proves to be extremely beneficial in terms of enhancing the collaboration that exists between the customer and the developer. In Agile development, estimation has always been difficult since requirements are subject to change. This motivates researchers to work on developing more accurate methods of effort estimation. The disparity between the amount of work that was estimated and the amount that was actually done. In this publication, a review was conducted of the work done by a large number of authors and potential researchers who were striving to close the gap between the real and estimated amount of effort. A thorough literature analysis led to the conclusion that machine learning models outperformed traditional estimation methods and methods that did not incorporate machine learning.

As humans, we can remember certain visuals in great detail, and sometimes even after viewing them once. What is even more interesting is that humans tend to remember and forget the same things, suggesting that there might be some general internal characteristics of an image that make it easier for the brain to encode and discard certain types of information. Research suggests that some pictures tend to be memorized more than others. The ability of an image to be remembered by different viewers is one of its intrinsic properties. In visualization and photography, creating memorable images is a difficult task. Hence, to solve the problem, various techniques predict visual memorability and manipulate images’ memorability. We present a comprehensive literature survey to assess the deep learning techniques used to predict and modify memorability. In particular, we analyze the use of convolutional neural networks, recurrent neural networks, and generative adversarial networks for image memorability prediction and modification.

Brain tumor is one of the deadliest diseases diagnosed in human beings. Doctors can identify the tumor with the Magnetic Resonance Imaging (MRI) technique which uses the magnetic field and computer-generated waves to create Brain images. With the advancement in the domain of Computer Vision, various researchers have come up with different state-of-the-art frameworks that aid doctors in diagnosing Brain tumors quickly. Most of these research works exploited the effectiveness of Convolutional Neural Networks (CNNs) on the image data. However, the major drawback of these studies is that they do not provide human-interpretable explanations for the models’ predictions. In this research work, the InceptionV3 CNN architecture is used to detect Brain tumors from the MRI images. This paper also provides human-interpretable predictions for Brain tumor diagnosis through the Local Interpretable Model-agnostic Explanation (LIME) framework to enhance the trust of doctors in the predictions of InceptionV3 CNN architecture. The MRI images of the Brain tumor have been gathered from the publicly available dataset catalog website Kaggle. InceptionV3 architecture is chosen as it is found to be the best architecture among the various CNN architectures, attaining an accuracy of 99.81%.

It is often the case that people set up elaborate goals but are not able to fully implement them. A “goal” is defined as the progress or familiarity with a topic in which the user wants to excel. There are applications which allow people to make and track goals and their progress in a particular domain, but are heavily dependent on the user’s input. Only the user knows whether they have completed their goal or not and the application is just reflecting what the user already knows. With Choice, a system would be introduced which not only lets people set their goals, but also monitors whether they are indeed sticking to them. The basis of the application is history distribution of a user. The application uses anonymous history tracking and Machine Learning to provide insights into search patterns, thus enabling better analysis of search topics for users. It is specifically targeted toward university students and can be coupled with a customizable chat-bot to further provide relevant and helpful information to the students.

In recent decades, significant research has been conducted in the field of Automatic Speech Recognition (ASR). Machine learning in the field of Natural Language Processing (NLP) is an innovative area of research and has become a popular topic for researchers. This paper aims to examine various Automatic Speech Recognition systems developed over the past decades. Antiquated ASR systems consist of acoustic model, language model, pronunciation model, Weighted Finite-State Transducer (WFST)-based decoder, and text-normalizer. These components are separately trained and assembled. We will discuss various end-to-end (E2E) structures such as Connectionist Temporal Classification (CTC), Recurrent Neural Network-Transducer (RNN-T), and the attention-based model. We will also explore some of the modern attention-based architectures such as Jasper, Transformer, and Conformer model.

Crowdfunding refers to investments of people in some projects and businesses. Sometimes, it also refers to donations from people for societal goodness. Many online platforms are available that support crowdfunding for different purposes. But most of the time investors or donors have a kind of doubt about whether their money is being utilized for the right purpose or not. The platforms are centralized; once the organization receives the donation, donors have no idea about the further transactions which the organization does. Many times, crowdfunding campaigns are not monitored and some of them have proven to be bogus. Due to such issues, projects have been extensively delayed in their completion. So, crowdfunding platforms are one area in which blockchain technologies can be used. Blockchain has been used in many different sectors to increase transparency among the users and people involved in the system. Blockchain is predicted to be used by the majority of technology as an effective method of conducting transactions online in the future. By integrating Ethereum-based contracts into the crowdfunding campaign, this project aims to overcome these problems, allowing the contracts to be fully automated, preventing fraud, and ensuring that projects are completed within the specified timeframe.

The process of assigning the correct meaning to a word with respect to its context is known as Word-Sense Disambiguation (WSD). It is a hard problem in natural language processing. In this paper, we discuss various approaches used to tackle this problem. Research works conduced in recent years to solve this problem are also discussed with special focus on resource poor languages. Technological trends followed in the recent years are analyzed that may help in the identification of future path to search better WSD solutions.

Alzheimer’s disease, a progressive neurological disorder, is one of the most common causes of dementia. This is one of the widely studied disorders to understand the changes in the brain and yet there is no cure. Having knowledge of various factors plays an important role in identifying this disease during its various stages of development. The aim of our work is to provide a system to identify the possibility of Alzheimer’s disease during its early stage of progress. This paper presents the analysis of different features of the case studies, as in demented and non-demented, to derive its relation and decide the category. Later the processed data is trained on machine learning models that can fit the data well. The final model will be able to provide a well-generalized hypothesis to classify a case as either likely to be demented or not.

Metaverse technology has gained much attention during the past few years. Especially during and after the pandemic, relevant technology is considered to provide greater accessibility and real-time, face-to-face experiences. This research also focused on metaverse technology, specifically in online advertising, further facilitating customer purchase decision-making. The researchers selected a sample of n = 384 individuals and employed structural equation modeling for the analysis. Results revealed that metaverse technology significantly affects online advertising, indicating its wider adoption among retailers in the United Arab Emirates. Besides, online advertising accompanied by metaverse technology significantly affects virtual simulation and observability. Finally, virtual simulation and observability significantly influence purchase decision-making, providing customers with real-time, face-to-face, 3D experiences. Overall, the results indicated a significant role of metaverse technology in the Emirati consumer industry, facilitating the customers to make the most suitable decisions. Thus, it is concluded that technology entices customers who enjoy these exposures and make decisions, further stressing the role and importance of metaverse in the consumer sector and underlining its potential role for retailers looking to grow their sales and customer loyalty by embracing technology.

Image data augmentation is the process of enlarging the dataset by creating synthetic images using the existing images to address the problem of scarcity of data and poor generalization. The existing methods for image data augmentation are mainly domain specific and focus on transformations in spatial domain. In this paper, we have developed a domain independent image data augmentation technique based on frequency domain transformations. In this research, we have utilized discrete cosine transform (DCT)-based latent space and proposed two techniques. In the first technique, we apply DCT on each of the color channel of the image and synthesize new image by considering only high-energy latent coefficients. The second technique is a hybrid technique where standard transformations are applied on the images synthesized from the DCT latent space. The experiments carried out on the Cifar10 dataset using VGG16 revealed that the proposed hybrid technique is more effective than the technique where the synthetic images are created solely from the DCT-based latent space. The analysis of training loss and validation loss shows significant improvement in validation loss, thus improving generalization of the VGG 16 model.

Artificial intelligence has significantly increased during the last ten years. AI-related text summarisation, which identifies the pertinent sentences from a text, is a significant research topic. Text summarisation may obtain concise and accurate information while maintaining the text’s original structure. The method for creating a succinct and accurate extraction summary for the provided text material is presented in this project. This paper provides a comprehensive literature review to gauge the complexities in the existing solutions, followed by the proposed solution using BERT for finding embeddings. As a classification challenge, extractive summarisation is done by predicting a class for each sentence in a document (i.e. determining if a sentence should or should not be a part of the generated summary). The sentences are assigned a rank, also known as their prediction score. The summary is then provided by ordering the best-scoring sentences according to certain relevant factors (e.g. order of appearance in the document, grammatical correctness, etc.). The news articles from CNN/Daily mail are used to verify the accuracy.

A word is treated as ‘ambiguous’ if it has more than one interpretation. Deciphering the meaning of an ambiguous word as per its context is known as word sense disambiguation (WSD). WSD, an open problem in natural language processing (NLP), has significant implication on various NLP applications, hence needs proper attention. Many efficient and robust approaches exist to tackle WSD issue and have been explored extensively for different languages. In this research work, the first attempt is made to analyze this problem in Kashmiri language. For this, we used two popular machine learning algorithms, k-NN and decision tree, and contrasted their performance. For experimentation, a novel sense tagged corpus is created for fifty commonly used and highly ambiguous Kashmiri words. The instances for the selected ambiguous words are extracted from a raw Kashmiri corpus of 500 K tokens. To tag the selected ambiguous terms with the appropriate sense, Kashmiri WordNet is used. Decision tree and k-NN-based classifiers are trained based on contextual features extracted from the sense tagged corpus. Both the algorithms are tested on all fifty target words. Accuracy, precision, recall and F1-measure are calculated for both the algorithms and compared. Although both the algorithms performed well, the decision tree-based classifier showed lower performance than the k-NN-based classifier in many cases.

Liver disease is one of the deadliest illnesses in the world. It takes place in the human body, particularly in the liver. The liver filters all of the blood present in the body and detoxicates harmful substances such as alcohol and drugs. Finding the origin and severity of liver disease is crucial for an effective treatment. The risk of liver disease can be predicted using various machine learning algorithms. We created a system that requests users submit the specifics of their blood test report based on the precise model. The system then uses the most precise model that has been trained to forecast whether or not a person is at risk for liver disease. Data preparation, data pre-processing, feature selection, classification, and building a model is the work flow which we followed. Indian_liver_patients and cirrhosis datasets are used. Mean, median, and standard deviation calculations are performed to enhance the input text data. The datasets consist of numerical and categorical data which is further pre-processed to remove the categorical and noise present within the dataset. Some unwanted features are eliminated using feature selection. With the aid of machine learning classification methods like Random Forest, Support Vector Machine, K-NN and Naive Bayes, liver illness can be detected early. The proposed work is an ensemble technique to predict the liver disease and its stages with higher accuracy when compared with individual classification techniques. Here we propose the combination of Random Forest, AdaBoost and GradientBoost classifiers. Concurrently, we use RandomizedSearchCV and GridSearchCV which are two effective ways that tune the parameters to increase the model generalizability. The performance metrics we considered are accuracy, precision, recall and F1-score.

A neurological disorder referred to as ASD autism spectrum disorder exists that may slow down speaking, linguistic abilities interpersonal skills. The symptoms typically occur within the initial period. Although ASD is mostly brought on by hereditary and external factors, it can occur from the time after newborn. Through the development phase, it can impact roughly one in every hundred people worldwide. Early detection and treatment can result in improvement; unfortunately, most kids with ASD do not acquire an accurate diagnosis and usually miss the opportunity for treatment; guardians may be hesitant to accept their child’s psychological progression which differs from one’s own physical growth. The delay in diagnosis prevents a toddler from receiving the necessary support and care to reach their full potential; at present, clinical standardised tests are the only available methods for diagnosing ASD. But they are time consuming, and costly efforts are underway to improve traditional procedures. Researchers have used approaches like support vector machines (SVM) and random forest classification methods (RFC) to build predictive model in order to enhance efficiency and accuracy. The study’s aim is to identify a child's vulnerability to ASD within the early phases aiding with early diagnosis. The study employed a systematic methodology to assess patient data over the previous 10 years who had disorder and non-chemical abnormalities. Findings demonstrate the effectiveness of using SVM and RFC, the RFC achieving 100% accuracies for all datasets. Early detection of ASD is crucial as larger amounts of information and testing can result in an increased accuracy for AI-assisted autism spectrum disorder which performs the exceptional among three arrangements for ordering ASD details.

Phishing is one of the most popular and hazardous cybercrime attacks. These attacks are designed to steal information used by people and companies to complete transactions. Phishing websites use a variety of indicators in their text and web browser-based data. This research presents a novel approach to classifying phishing websites by making use of the extreme learning machine (ELM). In this study, SVM, light GBM algorithm was used to detect phishing websites according to characteristics such as the length of their URLs, the number of capital letters they include and the presence of HTML elements. The findings indicate that ELM has a classification accuracy of 94.2% when it comes to phishing websites. This demonstrates the potential of ELM to classify websites that are used for phishing and to improve the safety of users who do their activities online.

The decision support predictive model for diabetes diagnosis is a valuable tool that can help healthcare professionals accurately predict diabetes outcomes and deliver the finest possible treatment to their patients. The main research purpose is to detect and classify the diabetes image by utilizing an ensemble learning method. The system uses a combination of techniques for feature selection and classification to detect the presence of diabetes, and we used SMOTE analysis to balance data from imbalanced data. The feature selection technique is used to identify the relevant factors that are associated with diabetes with the help of PSO. The model's performance is assessed utilizing measures like accuracy, recall, and F1-score. Results demonstrate the feasibility of recommended system in predicting a diabetes presence. The suggested system can be used as an effective decision support tool for early diagnosis and treatment of diabetes.

Stock market forecasting is fascinating yet challenging due to the influence of various factors. One such factor is sentiment which can influence the stock market movement. This paper considered historical stock data along with news sentiments to forecast stock market trends. The reason for considering sentiments is that the impact of news on the financial market cannot be ignored. Machine learning (ML) and ensemble learning are used since decades. Ensemble learning is quite useful in stock market forecasting but is complex and time consuming. Therefore, this paper proposes a less complex and faster ensemble learning approach which is based on multi-criteria decision-making (MCDM). The paper proposes Additive Ratio ASsement (ARAS)-based ensemble learning which uses the maximization and minimization criteria to select the optimal learning algorithm. For maximization criteria, accuracy (ACC), sensitivity (S), specificity (SP), and precision (P) are used, whereas for minimization criteria, false positive rate (FPR) and error rate (ER) are used. The proposed ARAS-based ensemble learning evaluates the best-performing model using the above six criteria from the performance metrics provided by the five popular ML algorithms. The proposed ensemble learning is tested on the two Indian stock indices (BSE and NIFTY50) where historical stock data of these indices are combined with the news sentiments obtained from the Times of India (TOI) news articles. The empirical outcome shows the superiority of ARAS-based learning over conventional ensemble learning approaches in terms of ACC and execution time (ET).

Multi-label learning affected from curse of dimensionality, therefore, feature selection is vital. In this paper, we propose feature ranking method using one of the well-known Multi Attribute Decision Making (MADM) technique. This method formulates Multi-Objectives Optimization by Ratio Analysis (MOORA) for multi-label feature selection. The proposed MOORA-Based Multi-Label Feature Selection (MBMLFS) uses Feature-Feature correlation as non-beneficial criteria and Feature-Label correlation as beneficial criteria to form the decision matrix. Person’s correlation and cosine similarity is used to compute correlation of features with features and labels, respectively. The MBMLFS uses decision matrix to rank the features. Six benchmark datasets (Bibtex, Birds, Corel5k, Scence, Enron, and Medical) are used to evaluate the MBMLFS. The MBMLFS is compared with five baseline feature selection methods. Results are statistically (Wilcoxon) tested to show the significance. The experiment shows that the proposed MBMLFS outperforms. The proposed MBMLFS is taking less execution time and won 83% of the time when compared to other well-known methods.

The rapid growth of the competitive market in today’s world has encouraged marketers to have more data-driven marketing strategies. As customers come from diverse backgrounds and have different requirements and expectations, understanding their demands and preferences is key to efficient customer relationships. This study shows the effect of different machine learning technologies in integration with the recency, frequency, and monetary (RFM) model and K-means to get meaningful customer segments. K-means is the most widely used method for customer segmentation because of its easy implementation, fast results, and accuracy. This study discovers the use of various extended RFMs in the literature for getting better segmentation results.

A group of wireless moving/still nodes establishes the mobile ad hoc network (MANET) that forms a temporary network. There does not exist any centralized access points, infrastructure, or centralized administration for controlling the network. Routing is the process of exchanging the data/control packets to achieve communication among moving nodes. Routing is the biggest challenge in MANET for finding a path between the communicating nodes. Swarm intelligence introduces various protocols based on the moving nature of living creatures to solve the optimization problems like routing. Ant colony algorithm is a specific type of subset of techniques of swarm intelligence in which simple ants have the capability to solve complex problems by the property of cooperation among them. One autonomous ant performs limited, specialized tasks, but the colonies of ants at large exhibit a global intelligent behavior. Stigmergy is the process of indirect communication shown by the ants for exchanging the data. Each ant continuously deposits chemical substance known by the name of pheromone on the path where ant traverses. Other ants have the capability to sense this chemical when they move over a particular path. The routing algorithm based on the ant’s behavior is well scalable, adaptive, and efficient in practice.

This paper mainly targets stress detection by analyzing the audio signals obtained from human beings. Deep learning is used to model the levels of stress pertaining to this whole paper followed by an analysis of the Mel spectrogram of the audio signals is done. A hybrid attention model helps us achieve the required result. The dataset that has been used for this article is the DAIC-WOZ dataset containing continuous speech files of conversations between a patient and a virtual assistant who is controlled by a human counselor from another room. The best results obtained were a 78.7% accuracy on the classification of the stress levels.

Language identification (LID) research is a significant area of study in speech processing. The construction of a language identification system is highly relevant in the Indian context, where almost every state has its language, and each language has many dialects. Social networking is becoming increasingly important in today’s social media platforms for people to convey their opinions and perspectives. As a result, it might be challenging to distinguish between specific languages in a multilingual nation like India. Data was gathered from publicly accessible Facebook postings and tagged with a code-mixed data tag created for this study. This study uses deep learning techniques to identify languages at the word level, where the encoded content may be in Hindi, Assamese, or English. Convolutional neural networks (CNNs) and long short-term memory (LSTM), two deep neural techniques, are compared to feature-based learning for this task. According to the finding, CNN has the best language identification performance with an accuracy of 89.46%.

Computer vision's most active area of research is handwritten digit recognition. Numerous applications essentially motivate to build of an effective recognition model using computer vision that can empower the computers to analyze the images in parallel to human vision. Most efforts have been devoted to recognizing the handwritten digits, while less attention has been paid to the recognition of handwritten digits in resource-poor languages such as Gujarati is a native language in India mainly due to the morphological variance in the writing style. We used a CNN-based model for digit recognition and a Convolution Neural Network for classification in this paper to propose a customized deep learning model for the recognition and classification of handwritten Gujarati digits. This work makes use of the 52,000 images in the Gujarati handwritten digit dataset, which was made from a scanned copy of the Gujarati handwritten digit script. The proposed method outperforms the current state-of-the-art classification accuracy of 99.17%, as demonstrated by extensive experimental results. In addition, the precision, recall, and F1 scores were calculated to assess the classification's effectiveness.

Cricket is the most played and accepted game in the Asian region of the world. In any game decision-making affecting the result plays a major role, which has to be accurate in every aspect to conclude the fair game. This matter is a live cricket match or even in training sessions. To date, decision-making was dependent on visual observation by the expert umpire. These manual observations of results may be inaccurate at some point and based on the experience and expertise of the umpire. With the rapid advancement and technological involvement in gaming industries, electronic gadgets are helping the player to enhance the performance and accuracy of results. This paper deals with the challenges and the possible solutions involved in designing and building a smart cricket ball, which will be capable to calculate and share the ball’s activity log. This log will involve the direction, speed, type, trajectory, spin, deviation angle, and impact triggers. It is difficult to get all information from a single type of sensor or module, this work suggests creating the fusion of multiple sensors and software algorithms for accurate measurements.

Heart disease is currently regarded as the main cause of illness. Regardless of age group, heart disease is a serious condition nowadays because most individuals are not aware of their kind and level of heart disease. In this fast-paced world, it is essential to be aware of the different types of cardiac problems and the routine disease monitoring process. As per the statistics from the World Health Organization, 17.5 million deaths are because of cardiovascular disease. Manual feature engineering, on the other hand, is difficult and generally requires the ability to choose the suitable technique. To resolve these issues, IoT, machine learning models and cloud techniques, are playing a significant role in the automatic disease prediction in medical field. SVM, Naive Bayes, Decision Tree, K-Nearest Neighbor, and Artificial Neural Network are some of the machine learning techniques used in the prediction of heart diseases. In this paper, we have described various research works, related heart disease dataset, and comparison and discussion of different machine learning models for prediction of heart disease and also described the research challenges, future scope and discussed the conclusion. The main goal of the paper is to review the latest and most relevant papers to identify the benefits, drawbacks, and research gaps in this field.

E-learning is recently days considered the easy media used to teach materials and courses. The E-learning environment contains a lot of resources and content related to each user profile. Those contents are considered private data or information for the user, so it is important to provide a secure environment for the users to keep their privacy safe. E-learning over fog computing takes place very fast, especially with the spread of (IoT) devices. Fog computing is offering a way to avoid latency and reduce the distance between the end users and resources over the cloud. This paper aimed to offer a secure method for the learning resources based on fog computing. Elliptic Curve Cryptography (ECC) algorithm is used to secure the data flow of the E-learning system over fog computing by using common and special key encryption in combination with AES. As the user data is encrypted and decrypted with the users’ keys to protect his privacy. The ECC is powerful encryption way and did not effects on the performance of the E-learning environment. At the end of the paper there is a comparison between RSA and ECC usage to illustrate the main difference between the two types of cryptography.

Smart environment is dominating today’s world, and this is leading to a surge in real-time applications for Wireless Sensor Network (WSN) technologies. But the downside of it are the issues concerning energy constraints as the energy of the battery-operated sensor nodes (SN) deplete with time. Clustering and routing approach is one of the techniques employed for increasing the energy effectiveness of SN. Multi-objective—Energy Centric Remora Optimization Algorithm (MO-ECROA) proposes to perform cluster-based routing by using various fitness functions for the purpose of best candidate selection for cluster head (CH). It then uses Ant Colony Optimization Algorithm (ACO) for choosing the optimal routing path for forwarding packets to the Base Station (BS) from CHs. The proposed MO-ECROA, moderates the node energy consumption while enhancing WSN data delivery. It shows better performance as far as energy efficiency and network throughput is concerned. The energy effectiveness of the MO-ECROA method-based CH protocol is 82.64%, which is higher when compared to the existing ACI-GSO, MWCSGA, and O-EHO.

Since the corona virus was discovered, there has been an increase in the difficulty of gaining access to genuine clinical resources. This includes a scarcity of experts and healthcare workers and an absence of appropriate equipment and drugs. The whole medical community is in a state of crisis, which has led to the deaths of a significant number of people. Because the drug was not readily available, people began self-medicating without first consulting with their doctors, making their health situation much worse. Recently, machine learning has shown to be helpful in a wide variety of applications, and there has been an uptick in the amount of new work done for automation. The study’s goal is to showcase a medicine recommender system (RS) that can drastically cut down on the amount of labor now being done by specialists. In this research, we build a drug recommendation system by analyzing patient feedback for tone. We utilize the machine learning-based XGBoost classifier, count vectorization for feature extraction, and ADASYN for data balancing. This system can assist in recommending the best drug for a specific disease by using a variety of implementation processes. The predicted sentiments were established based on their precision, recall, accuracy, F1-score, and area under the curve (AUC). The findings suggest that 95% accuracy may be achieved using the classification algorithm XGBoost with count vectorization compared to other models. The results of our experiments demonstrate that our system can provide highly accurate, efficient, and scalable drug recommendations.

With the increased security benefits provided by blockchain, it has made its place in domains other than cryptocurrency, namely, Internet of Things (IoT), supply chain, health care, Unarmed Aerial Vehicles (UAVs), mobile ad hoc networks (MANETs), and what not. In today's technologically advanced world, cryptographic key management is an essential component of communication security. An efficient and secure key management is a challenge for any cryptographic system because an intruder could be present anywhere in the system. This paper has visualized a novel concept where blockchain-based different key management schemes have been studied across five domains and this is one of the first efforts in this manner. Several schemes have been proposed in the past and it is further being research upon because a particular scheme is not profound to every domain when it comes to security, performance, and scalability challenges.

Visual Arts reflect the knowledge gained by humans within an era. These arts assist in insight into the progress in the sphere of culture, history, and lifestyle. In recent years, the digitization of Indian Visual Arts has initiated the promotion of tourism across the globe. The digitization is producing repositories. These repositories can be explored to accomplish different research as classification. Now, moment comes to do work for classification of the Indian Visual Arts. The classification can be performed using different models of Neural Networks. Two models of Neural Networks, AlexNet and CNN fc6, were used in the study. The results of models were investigated based on the performance evaluation matrices classification accuracy, precision, recall, and F1-score. The classification accuracy for the AlexNet model was best among both models. For AlexNet, the classification accuracy, precision, recall, and F1-score were seventy-two percent, seventy-one point eight eight, sixty-nine point six seven, sixty-seven point nine nine, respectively.

Underwater communications are playing a major role in recent times as underwater happenings have gained interest. Especially in scenarios based on natural calamities like tsunamis, pollution detection, and surveillance, underwater communications are indispensable. Parameters like battery life span and capabilities of sensor nodes play a vital role in the underwater environment. QoS has to be taken care of in underwater communication. Congestion avoidance helps in using energy efficiently in underwater communication. Simulation results prove that this algorithm overcomes the drawback of a normal backpressure-based, static priority algorithm where packet loss, delay, the energy of nodes, and routing overhead are reduced overall, and packet delivery ratio and throughput are increased in special cases of real-time important packets. As underwater nodes are battery-operated, it is essential to have energy-efficient scheduling algorithm. The proposed algorithm solves the drawback of the backpressure algorithm by having energy-efficient queue length stabilizer (QLS) technique, where the backlog is adjusted such that no queues are affected and the energy consumption is less. And real-time packets are treated effectively such that real-time applications can be supported using this scheduling algorithm. The novelty of the proposed method is congestion control methodology with active priority and energy efficiency for an underwater network for effective scheduling of packets where queue backlog size is also adjusted virtually based on energy.

Identifying which part of the image has been manipulated in real-world photo is a challenging research area. Generally, images are tampered using image editing tools to hide the information from pictures and to mislead people by providing wrong information intentionally. The authenticity of images becomes crucial to check for whether the image has been tampered before extracting meaningful information from images. In his paper, copy-move forgery area detection, one of several tampering techniques, has been presented based on identifying Scale-Invariant Feature Transform (SIFT) keypoints in an image. Preprocessing is done using fuzzy contrast enhancement. Features are extracted using SIFT algorithm and identical portions are clustered using Density-Based Spatial Clustering Application with Noise (DBSCAN) clustering. The experiments are performed on MICC-F220 dataset and results obtained are compared with other methods in the literature.

A dual-band slot-coupled MIMO antenna has been designed for base station applications. The radiating structure consists of eight antennas placed on top of Rogers RT/Duroid substrate of permittivity 2.2 and thickness 1.6 mm with four antennas operating under transmit mode and four operating under receive mode. By making U shaped slot on the antenna structure resulted in dual resonant frequencies corresponding to 28 GHz and 36 GHz. The novelty of the proposed work lies in using a rectangular annular ring loaded with dumbbell-shaped defected ground structure (DGS) etched on the ground plane of the antenna to increase the bandwidth and reduce the coupling. Further, square-shaped planar electromagnetic bandgap (EBG) cells have been added to improve the gain of the antenna array. The coupling between the antenna elements of the array decreased from −10 dB to −15 dB and the response remained flat in the operating bandwidth of the antenna with the introduction of DGS. The structure resulted in an envelope correlation coefficient (ECC) of 0.003 and a diversity gain of 10dB in the operating band of the antenna structure.

IoT is a dynamic, emerging sector characterized by using existing Internet infrastructure to connect devices and data centers worldwide. The internet of healthcare things (IoHT) is a subclass of IoT that includes high-tech medical equipment that plays a crucial role in data monitoring, processing, storage, and transfer. It has novel challenges in data security. Several cryptographic techniques have been devised to safeguard the system against data misuse, alteration, and node tempering. Such cryptographic algorithms are inadequate because of the device's diminutive size, limited computing power, memory, and power resources. A lightweight cryptographic scheme based on symmetric cryptosystems powered by the Feistel structure is required to secure such a system. A lightweight cryptosystem is proposed in this paper for the security of the e-healthcare system. The proposed algorithm encompasses a compact S-box function, less round and addition substitution, and an XOR operation to generate encrypted information from plaintext. Additionally, a secure authentication mechanism based on the previously mentioned method is suggested for secure communication in the healthcare system. The cryptanalysis process and result state that our proposed system resists several cryptanalytic attacks which include brute force, avalanche effect, linear cryptanalysis, and biclique attack when compared with standard AES algorithm.

Design studio (DS) as teaching approach finds its applications in curriculum design, and DS environments support students of different disciplines to follow sequential curriculums to develop their skills and learning patterns. Visualization of knowledge offers researchers a visual viewpoint and aids in comprehending the overall design structures of certain study topics more effectively. This study presents a systematic bibliometric analysis of the knowledge dynamics and structure, and micro- and macro-hotspots of DS research using VOSviewer. The study covers articles published in journal and conference indexed in the Scopus Database between 1974 and 2022 and implements visualization method to quantify the numbers of articles, referenced sources, cited documents, top-publishing organizations, and production houses in countries focused to outsource research on DS. The study findings would guide scholars and researcher, in particular instructional designers, interested in expanding and broadening the scope and horizon of research concerning DS.

The sensor node is a key component of a WSN. The battery life of a sensor node determines how well it operates. Once batteries are placed in a remote or unattended location, it is proven to be impossible to replace them. In this paper, to increase the lifespan of a network, an improved optimal cluster head (CH) selection method based on grid-based whale optimization (GBWOA) is proposed. It uses the optimized CH selection in the specified grid area of the network, and the integration of numerous factors for the CH selection is a novel feature of the presented algorithm. Energy consumption, throughput, and network lifetime are used to assess how well the suggested algorithm performs. When network lifespan is compared to the current approach, the suggested approach outperforms the whale optimization algorithm (WOA-C) and particle swarm optimization (PSO-C) by 22% and 71%, respectively.

As a result of their inherent energy constraints, Wireless Sensor Networks (WSNs) researching energy-efficient routing protocols is a top priority. Therefore, it is crucial to make the most of the available power to lengthen the operational lifespan of WSNs. This work introduces a 200 m2 field implementation of a modified Energy-Efficient Multi-hop Routing Protocol (mEEMRP). The protocol depends on such a technique for distributing load among Communication Management (CM) networks using multi-hop routing of the available information toward the Base Station (BS) while taking into account the Residual Energy (RE) values for CM nodes and the range with both neighboring CM nodes. Based on simulation findings, mEEMRP outperformed an efficient energy member of the co-routing protocol (EEMRP) in terms of the network lifespan by 1.77%. Further, the suggested mEEMRP reduced BS energy usage by 4.83% and increased packet reception by 7.41%.

Automatic recognition of species is the task of identifying and counting animal or bird species from pictures taken from camera traps. Such recognition systems help ecologists automatically analyse and monitor animal behaviour without human intervention. In this work, we exploit transfer learning using convolutional neural networks (CNN) to identify animal species. The overall framework uses a pre-trained network as a backbone to learn general features before the classification layer. Using this framework, several models are developed by using the EfficientNet, ResNet, Inception, and VGG as the backbone networks. Each model is trained over the animal species dataset. The models are evaluated over test data, and it is observed that the EfficientNet-based model exhibits the best performance.

Nowadays, parents are concerned about their children due to the increased occurrence of kidnapping. At the same time, they just do not have as much time to spend with their children due to their heavy workloads in their offices. As a result, children are vulnerable to being convinced by kidnappers before entering the school. Students feel safer at school, and parents have more faith in educational institutions. This article discusses a student location monitoring system built on an Arduino Nano that uses GSM and GPS technologies to track the student using GPS tracking, which improves security and safety for the student. A security perimeter is established around the school’s grounds using geofencing technology, with the school as its central point. It records student arrival and departure times from the school grounds and sends SMS immediately to their parents, confirming that the student arrived at school safely. It also sends SMS notifications to their parents once their children leave the school. The establishment of a student security system and student monitoring using GPS tracking is to prevent crime and illegal activities by pupils and alleviate parental anxieties.

Gastrointestinal diseases are increasing at a fast rate. Some of these lead to colorectal cancer. The presence of polyps in the large intestine may lead to colorectal cancer in later stages. Early detection and prediction of colorectal cancer is very crucial as it is the third most occurring cancer in the world. In this study, different deep learning methods for image classification were implemented to classify various gastrointestinal diseases including polyps detection. The ResNet50 model implemented with transfer learning achieved classification accuracy of 99.25% on training set. The EfficientNet model achieved classification accuracy of 93.25% on validation set and 94.75% on test set.

Internet of Things is a popular source to collect data. It is also a rich source to various types of information. With the rapid popularity of IoT, things getting connected to it and the number are increasing continuously. Hence, the challenge associated with the proper maintenance of IoT networks in different sectors is increasing globally and the growing size is the ultimate reason for this problem. DDoS is one of the various attacks which are common and known. Botnets are being used to perform such attacks. Machine learning is a technology that has been supporting the standard computing environment in many ways. It can help design efficient models to identify attacks. Recent standard datasets and machine learning techniques, such as, Decision Trees, Random Forest, and KNN, are used in this work to ascertain DDoS attacks performed on IoT environments. These methods are compared by considering the confusion matrix created on the basis of different measures.

Online comments can often be toxic, offensive, and harmful to individuals and communities. In recent years, there has been a growing need to automatically identify and mitigate these toxic comments. For this problem, NLP models are often used to identify such toxicity and harshness but each model has its own efficiency and performance limitations. In this paper, we propose the use of the bidirectional encoder representations from transformers (BERT) algorithm for toxicity classification of online comments. BERT is a state-of-the-art natural language processing model developed by Google in 2018 that has shown strong results on a variety of tasks. In this paper, we used the BERT algorithm for toxicity classification and evaluated its performance on a real world dataset and performed comparative analysis with conventional NLP models, logistic regression (TF-IDF) over which BERT showed an improvement of 6.9% in accuracy, 26.1% in f1-score, 21.5% in ROC score; logistic regression (BOW) over which BERT showed an improvement of 9.1% in accuracy, 70.6% in f1-score, 39.8% in ROC score; multinomialNB (BOW) over which BERT showed an improvement of 9.2% in accuracy, 25.9% in f1-score, 10.6% in ROC score.

Classification accuracy for imbalance class data is a primary issue in machine learning. Most classification algorithms result in insignificant accuracy when used over class imbalance data. Class imbalance data exist in many sensitive domains such as medicine, finance, etc., where infrequent events such as rare disease diagnoses and fraud transactions are required to be identified. In these domains, correct classification is essential. The paper presents a hybrid sampling model called KSMOTEEN to address class imbalance data. The model uses a clustering approach, the K-means clustering algorithm, and combines the SMOTEEN technique. The experimental result shows, the KSMOTEEN outperforms some existing sampling methods, thus improving the performance of classifiers for class imbalance data.

Protocol conformance testing is an indispensable part of the marketization of commercial terminals. This paper elaborates and studies the architecture, test model, and design scheme of the 5G terminal protocol conformance testing system and proposes a 5G terminal protocol based on TTCN-3 Conformance test codec solution. The outcome of research shows that this solution has a positive role in promoting the industrialization of 5G.

Transportation is necessary for modern living, yet the conventional combustion engine is quickly going out of style. All electric vehicles are quickly replacing gasoline and diesel vehicles because they create less pollution. The environment is greatly improved by fully electric vehicles (EVs), which produce no exhaust pollutants. Using modelling and optimization, researchers have concentrated on building smart scheduling algorithms to control the demand for public charging. To develop better forecasts, consider aspects such as prior historical data, start time, departure time, charge time hours, weekday, platform, and location id. Previous research has used algorithms like SVM and XGBOOST, with session time and energy usage receiving SMAPE ratings of 9.9% and 11.6%, respectively. The classifier model in the suggested method which makes use of CNN sequential architecture achieves the best prediction performance as a consequence. We emphasize the importance of charging behaviour predictions in both forecasts relative to one another and demonstrate a notable advancement over earlier work on a different dataset. Using various lengths of training data, we assess the behaviour prediction performance for increasing charge duration levels and charging time slots in contrast to prior work. The performance of the proposed technique is verified using actual EV charging data, and a comparison with other machine learning algorithms shows that it generally has higher prediction accuracy across all resolutions.

Today, one of the emerging challenges faced by neurologists is to categorize Alzheimer's disease (AD). It is a type of neurodegenerative disorder and leads to progressive mental loss and is known as Alzheimer's disease (AD) (Tanveer et al. in Commun Appl 16:1–35, 2020). An immediate diagnosis of Alzheimer's disease is one of the requirements and developing an effective treatment strategy and stopping the disease’s progression. Resonance magnetic imaging (MRI) and CT scans can enable local changes in brain structure and quantify disease-related damage. The standard machine learning algorithms are designed to detect AD to have poor performance because they were trained using insufficient sample data. In comparison with traditional machine learning algorithms, deep learning models have shown superior performance in most of the research studies stated specific to diagnosis of AD. One of the elegant DL method is the convolutional neural network (CNN) and has helped to assist the early diagnosis of AD (Sethi et al. in BioMed Research International, 2022; Islam and Zhang in Proceedings IEEE/CVF 841 conference computing vision pattern recognition workshops (CVPRW), pp 1881–1883, 2018). However, in recent days advanced DL methods have also attempted for classification of AD, especially in MRI images (Tiwari et al. in Int J Nanomed 14:5541, 2019). The purpose of this paper is to propose a ResNet50 model for Alzheimer's disease, namely AD-ResNet50 for MRI images that incorporates two extensions known as transfer learning and SMOTE. This research uses the proposed method and compares it with the standard deep models VGG19, InceptionResNet V2, and DenseNet169 with transfer learning and SMOTE (Chawla et al. in J Artif Intell Res 16:(1)321–357, 2002). The results demonstrate the efficiency of the proposed method, which outperforms the other three models tested. When compared with baseline deep learning models, the proposed model outperformed them in terms of accuracy and ROC values.

Although air quality prediction is a crucial tool for weather forecasting and air quality management, algorithms for making predictions that are based on a single model are prone to overfitting. In order to address the complexity of air quality prediction, a prediction approach based on integrated dual long short-term memory (LSTM) models was developed in this study. The model takes into account the variables that affect air quality such as nearby station data and weather information. Finally, two models are integrated using the eXtreme Gradient Boosting (XGBoosting) tree. The ultimate results of the prediction may be obtained by summing the predicted values of the ideal subtree nodes. The proposed method was tested and examined using five evaluation techniques. The accuracy of the prediction data in our model has significantly increased when compared with other models.

This paper designs an epidemic prevention and control mask wearing detection system based on STM32, which is used to monitor the situation of people wearing masks. Tiny-YOLO detection algorithm is adopted in the system, combined with image recognition technology, and two kinds of image data with and without masks are used for network training. Then, the trained model can be used to carry out real-time automatic supervision on the wearing of masks in the surveillance video. When the wrong wearing or not wearing masks are detected, the buzzer will send an alarm, so as to effectively monitor the wearing of masks and remind relevant personnel to wear masks correctly.

As customer traffic has been increasing over the years on online shopping websites, it is indispensable for sellers to assess online customers’ purchase intentions, which can potentially be predicted by analyzing the historical activities of the customers. This study analyzes the highly imbalanced empirical data of online shoppers’ intentions to foretell whether a visitor to an online shopping website will make a purchase. The synthetic minority oversampling technique has been implemented to reconstruct the dataset to alleviate the class imbalance in the original dataset. The effectiveness of oversampling has been identified by comparing the predictive performance of four different classifiers Partial decision tree (PART), decision tree (DT), Naïve Bayes (NB), and logistic regression (LR) on the reconstructed data with the performance on the original dataset. It has been observed that each classifier performs better on the reconstructed dataset. Ensemble learners have been implemented with varying base classifiers on the reconstructed dataset to identify the best predictive model. Bagging, boosting, and max-voting ensemble learners have been implemented with the base classifiers PART, DT, NB, and LR. The best performance has been observed by the prediction using bagging with PART as the base classifier with an accuracy of 92.62%. Hence, it has been identified as the best model for predicting the purchase intention of a customer in terms of accuracy. However, the highest precision and recall values of 0.923 have been given by the max-voting classifier with DT, PART, and LR as the base learners. It has also been concluded that the proposed methodology outperforms the existing models for shoppers’ intention section tasks.

With the ever so growing internet, its influence over the society has deepened, and one such example is social media as even children are quite active on social media and can be easily influenced by it, social media can be a breeding ground for cyberbullying, which can lead to serious mental health consequences for victims. To counter such problems, content moderation systems can be an effective solution. They are designed to monitor and manage online content, with the goal of ensuring that it adheres to specific guidelines and standards. One such system based on natural language processing is described in the following paper, and various algorithms are compared to increase accuracy and precision. After testing the application, logistic regression yielded maximum precision and accuracy among the other algorithms.

Automated cars are being developed by leading automakers and this technology is expected to revolutionize how people experience transportation, giving people more options and convenience. The implementation of traffic signal detection and recognition systems in automated vehicles can help reduce the number of fatalities due to traffic mishaps, improve road safety and efficiency, decrease traffic congestion, and help reduce air pollution. Once the traffic signs and lights are detected, the driver can then take the necessary actions to ensure a safe journey. We propose a method for traffic sign detection and recognition using ensemble techniques on four models, namely BEiT, Yolo V5, Faster-CNN, and sequential CNN. Current research focuses on traffic sign detection using an individual model like CNN. To further boost the accuracy of object detection, our proposed approach uses a combination of the average, AND, OR, and weighted-fusion strategies to combine the outputs of the different ensembles. The testing in this project utilizes the German Traffic Sign Recognition Benchmark (GTSRB), Belgium Traffic Sign image data, and Road Sign Detection datasets. In comparison with the individual models for object detection, the ensemble of these models was able to increase the model accuracy to 99.54% with validation accuracy of 99.74% and test accuracy of 99.34%.

The loss of customers is a serious issue that needs to be addressed by all major businesses. Companies, especially in the telecommunications industry, are trying to find ways to predict customer churn because of the direct impact on revenue. Therefore, it is important to identify the causes of customer churn to take measures to decrease it. Customer churn occurs when a company loses customers because of factors such as the introduction of new offerings by rivals or disruptions in service. Under these circumstances, customers often decide to end their subscription. Predicting the likelihood of a customer defecting by analyzing their past actions, current circumstances, and demographic data is the focus of customer churn predictive modeling. Predicting customer churn is a well-studied problem in the fields of data mining and machine learning. A common method for dealing with this issue is to employ classification algorithms to study the behaviors of both churners and non-churners. However, the current state-of-the-art classification algorithms are not well aligned with commercial goals because the training and evaluation phases of the models do not account for the actual financial costs and benefits. Different types of misclassification errors have different costs, so cost-sensitive learning (CSL) methods for learning on data have been proposed over the years. In this work, we present the CSL version of various machine learning methods for Telecom Customer Churn Predictive Model. Furthermore, also adopted feature selection strategies along with CSL in real-time telecom dataset from the UCI repository. The proposed combination of CSL with ML, the results outperforms the state-of-the-art machine learning techniques in terms of prediction accuracy, precision, sensitivity, area under the ROC curve, and F1-score.

In recent days, for security services, an intrusion detection system (IDS) is a highly effective solution (Louk MHL, Tama BA. “PSO-Driven Feature Selection and Hybrid Ensemble for Network Anomaly Detection,” Big Data and Cognitive Computing, 2022; Chebrolu et al. in Comput Secur 24:295–307, 2005). The primary goal of IDSs is to facilitate the detection of sophisticated network attacks by empowering protection instruments. Multiple ML/DL algorithms have been proposed for IDS (Choi H, Kim M, Lee G, Kim W. Unsupervised learning approach for network intrusion detection system using autoencoders. The Journal of Supercomputing, 2019). In this work, proposed IDS-PSO-BAE, an ensemble framework to improve the performance of the IDS using PSO-based feature selection, and bagging-based autoencoder classification. The hyperparameter settings of the autoencoders result in the best detection performance, and this method is good for unknown types of attacks’ detection. With bagging, a weak learner can be transformed into an effective learner, enabling precise classification. The best set of features to serve into the ensemble model is selected using PSO-enabled feature selection. The study, in which the complete train dataset is split into set of sub data sets and applied autoencoder on individual intrusion subset with specific ensemble learning. At the end, all results of individual ensemble learning are combined as final class prediction with the voting technique. The final feature subsets from the NSL-KDD dataset (Dhanabal and Shantharajah in Int. J. Adv. Res. Comput. Commun. Eng. 4–6:446–452, 2015) are trained with a hybrid ensemble learner for IDS. The results of the IDS-PSO-BAE model have superior accuracy, recall, and F-score compared to standard methods.

The Internet of Things continues to grow in size, connection, and applicability. Just like other new technologies, this ecosystem affects every area of our daily existence (Kafle et al. in IEEE Commun Mag 54:43–49, 2016). Despite the many advantages of the Internet of Things (IoT), the importance of securing its expanded attack surface has never been higher. There has been a recent increase in reports of botnet threats moving into the Internet of Things (IoT) environment. As a result, finding effective methods to secure IoT systems is a critical and challenging area of study. Potentially useful alternatives include methods based on machine learning, which can identify suspicious activities and even uncover network attacks. Simply relying on one machine learning strategy may lead to inaccuracies in data collection, processing, and representation if applied in practice. This research uses stacked ensemble learning to detect attacks better than conventional learning, which uses one algorithm for intruder detection (ID). To evaluate how well the stacked ensemble system performs in comparison to other common machine learning algorithms like Decision Tree (DT), random forest (RF), Naive Bayes (NB), and support vector machine (SVM), the BoT-IoT benchmark dataset has been used. Based on the findings of the experiments, stacked ensemble learning is the best method for classifying attacks currently available. Our experimental outcomes were assessed for validation data set, accuracy, precision, recall, and F1-score. Our results were competitive with best accuracy and ROC values when benchmarked against existing research.

Recommender systems (RS) have been around us since the beginning of the new age of technology, consisting of artificial intelligence, machine learning, the Internet of things (IoT), etc. The RS provides a personalized touch to the customers, thus helping them in decision-making. It also helps the business improve sales; hence, big tech companies like Netflix, amazon, etc. rely hugely on their RS to gain more sales. Many studies are focused on improving the accuracy of the RS but little on the security aspects of RS. Blockchain technology is the epitome of security and privacy. Hence, this research focuses on integrating blockchain with recommender engines and their advantages and challenges.

Every day, thousands of research papers are produced, and amongst all of these research works, computer science is most continually evolving. Thus, a large number of academics, research institutions, and funding bodies benefit from knowing which research fields are popular in this specific field of study. In this regard, we have produce a deep learning-based framework to estimate the future paths of computer science research by forecasting the number of articles that will be published. The recommended strategy shows the best prediction results in contemporary to the baseline approaches with 1483.23 RMSE and 0.9854 R-Square values.

Modern farming practices such as precision agriculture make crop production more efficient. The earlier detection of plant diseases is one of the major challenges in the agricultural domain. There is currently a substantial time and accuracy gap between manual plant disease classification and counting. In order to prevent the damage that may be caused to plants, farmers, and the agricultural ecosystem at large, it is essential to detect different diseases of plants. The purpose of this project was to classify and detect plant diseases, especially those that affect tomato plants. We propose a deep convolutional neural network-based architecture for detecting and classifying leaf disease. Images from Unmanned Aerial Vehicles (UAVs) are used in the experiment. There is also a dataset of plant village images, a dataset of UAV images, real-time UAV images, and an image from the internet used as well. Hence, detection of diseases is done with more accuracy as multiple datasets are used.

Traffic rule violations and accidents are major sources of inconvenience and danger on the road. In this paper, we propose a convolutional neural network (CNN)-based approach for detecting these events in real-time video streams. Our approach uses a YOLO-based object detection model to detect vehicles and other objects in the video and an IOU-based accident detection module to identify potential accidents. We evaluate the performance of our approach on a large dataset of traffic video footage and demonstrate its effectiveness in detecting traffic rule violations and accidents in real time. Our approach is able to accurately detect a wide range of traffic rule violations, including wrong-side driving, signal jumping, and over-speed. It is also able to accurately track the movements of objects in the video and to identify potential accidents based on their trajectories. In addition to detecting traffic rule violations and accidents, our approach also uses an ANPR module to automatically read the license plate numbers of detected vehicles. This allows us to generate e-challans and punishments for traffic rule violations, providing a potential deterrent to future violations. Overall, our proposed approach shows promise as a tool for detecting and preventing traffic rule violations and accidents in real-time surveillance systems. By combining powerful object detection and motion analysis algorithms with an ANPR module, it is able to accurately and efficiently identify traffic rule violations and accidents, providing valuable information for traffic management and safety.

Plant disease infestation causes severe crop damage and adversely affects crop yield. This damage can be reduced if diseases are identified in the early stages. Initially, farmers and agricultural scientists used to diagnose plant diseases with their naked eyes. With the dawn of different advanced computer vision techniques, various researchers have utilized these techniques for automatic disease detection in plants using their leaf images. In this research work, a novel triple attention embedded vision transformer is proposed for automatically diagnosing diseases in plants. In the proposed model, channel and spatial attention are embedded in addition to the multi-headed attention of the original vision transformer model. The reason for embedding the channel attention and spatial attention in vision transformer is that the existing multi-headed attention of vision transformer only considers the global relationship between the features and ignores the spatial and channel relationship. Moreover, in order to increase the confidence of farmers and agricultural scientists in predictions of the proposed model, human-understandable visual explanations are also provided with the predictions. These explanations are generated using the local interpretable model-agnostic explanations (LIME) framework. The experimentation of this research work is carried out on one publicly available dataset (PlantVillage dataset) and one real-in-field dataset (Maize dataset) having complex background images. For each dataset, it is experimentally found that the proposed model outperformed other research works found in the literature. Moreover, the visual explanations for the predictions of the proposed model highlight the infected area of leaves for diseased class predictions.

Cyber security crimes continue to increase every day. As the devices and the network connectivity is increasing, attackers and hackers committing the crimes over these diversely connected devices are also increasing. This brings a major attention to stop these attacks, and the focus has been moved to machine learning cyber threats, but with its advancement, it has now been used in multiple different ways to reduce the cyber-attacks. Although the non-availability of proper dataset is one of the limitations which were there in most of the studies. This paper will give the extensive details about the research done to understand the ML models and to use for those in preventing cyber-attacks models to protect the devices from the three major domains of attacks, namely spam, malware, and intrusions attacks. This paper will present the review of the studies which have been done previously to reduce the cyber-attacks using machine learning. It will discuss the implementation of the commonly used models used to predict the intrusions, malware, and spam detection, followed by which there will be a comparative analysis of these models for the three domains of the cyber-attacks. This review will also discuss the limitations and the future work to enhance the security of the network devices.

Autism, frequently called autism spectrum disorder (ASD), is considered a chief developmental illness impacting people's ability to talk and socialize. It contains a huge variety of troubles marked by difficulties with communication skills, repeated behaviors, speech, and nonverbal communication. The goal of this study was to develop a low-cost, quick, and simple autism detector. Based on a brief, structured questionnaire, the model provided in this investigation was trained to diagnose autism. The questionnaire consists of ten yes or no questions, each of which was connected to the everyday life of an autistic patient. The model identifies whether the user had ASD based on the responses provided by the users. Over the dataset “autism screening adult dataset”, machine learning algorithms such as Naive Bayes (NB), Classification and Regression Tree (CART), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) classifier were used. The most suitable models were found SVM and LDA with an accuracy percentage of 87.07%.