Advances in Computing and Data Sciences

Using the DEAP dataset, this conference paper aims to investigate the effectiveness of K-Nearest neighbours (KNN) and Multilayer Perceptron (MLP) classifiers in the context of emotion recognition. The primary focus is on decoding valence-arousal emotions, with special attention to how EEG bands are represented in relation to time and channels. The study comprises pre-processing and feature extraction from the multimodal DEAP dataset, which contains EEG signals associated with emotional responses. The Valence-Arousal model is chosen because it may be used to capture the essential elements of affective experiences. KNN and MLP, two distinct classifiers, are used to assess how well they distinguish emotions from EEG signals. The efficacy of these classifiers is evaluated using a range of metrics, such as accuracy, precision, re-call, and F1-score, which offer a thorough grasp of both their benefits and drawbacks. The study also examines the representation of EEG bands across time and across channels in order to find trends and connections in emotional responses. This means taking a close look at the ways in which different frequency bands help distinguish between Valence-Arousal emotions and offer insights into the temporal and spatial dynamics of emotional processing. The EEG data processing process employed the Multi-layer Perceptron (MLP) and k-Nearest neighbours (KNN) algorithms to assess the precision of the Arousal and Valence classifications in different regions of the brain. In the left region, KNN outperformed MLP in terms of arousal (69.11% vs. 60.16%) and valence (69.35% vs. 62.89%). Similarly, KNN scored better than MLP in the parietal region in terms of accuracy for valence (66.67%) and arousal (68.29% vs. 60.16%). In the middle region, KNN outperformed MLP in terms of valence accuracy (72.34% vs. 65.77%) whereas MLP had a lower arousal accuracy (57.72% vs. 68.29%). These results emphasise how crucial it is to take into account technique choice as well as brain regions when evaluating EEG data related to affective states. The findings may have an impact on the creation of more accurate and efficient emotion recognition systems, which may have an impact on applications in affective computing, human-computer interface, and medicine.

Software development environment setup and configuration is a very tedious and time consuming task for any organization. It is challenging for those peoples who are not technically competent to configure the development environment. Delay in setting up development environment may result, loss in productivity and revenue. To overcome these problems, we proposed a Docker based computing environment which provides a quick development environment to the end users. The proposed architecture speeds up the procedure for setting up a development environment by enabling the users to start a tech stack of their choice with few clicks. The proposed system keeps their users away from the necessity for complex local system installation and provides high performance computing environment like a mainframe.

Since sentiment analysis has been a crucial field for various types of studies, this field needs to develop and improvise its efficiency and accuracy. The problem has been analyzed over time in many different ways, and multiple approaches have been used to solve it. So far, we haven’t utilized the combined power of modern models and contextual information obtained from texts, nor has it analyzed how contextual information could be effectively controlled. By combining deep learning models with rule-based techniques, the proposed approach effectively classifies sentiments for a range of topics. Remarkably, it uses an embedded representation and attention mechanism to process valence shifting cases efficiently. The method, that employs sentiment dictionaries related to specific domains and custom-designed rules, shows better results on three datasets. Probability Proportion Difference (PPD) and Categorical Probability Proportion Difference (CPPD) are used in feature selection, these methods outperformed Information Gain and the standard method. Comparable sentiment detection abilities are revealed through a comparative analysis with Categorical Probability Difference (CPD). Tests performed on datasets shows the superior sentiment classification performance of the CPPD method over other methods.

This study investigates the performance of Julia and compares it with the C and Python programming languages in solving problems involving one-dimensional vectors, matrix operations, and two-dimensional equations. Through a series of experiments, Julia’s performance on different computers with different numbers of CPU cores is measured and compared. We evaluated Julia’s performance using numerical symbols for one-dimensional vector and matrix equations on a single switch, and then tested Julia’s MPI implementation using 2D thermal equations in a multi-core Compute environment. The results show that Julia offers the best performance, often compared to C and better than Python, especially on tasks that require a lot of computation, such as one-dimensional vector operations and matrix multiplication. Comparison of the performance of C and Julia is illustrated by the simultaneous use of parallelism on both sides; This demonstrates the Julia language’s ability to use parallelism. Julia’s performance in performing computational tasks demonstrates her aptitude for computational science and statistical analysis.

A malware can completely destroy the confidentiality, integrity, and availability of a system once it comes into contact with it, it poses a serious threat to computer security. Various mitigation and detection techniques have been developed over decades to address this problem. In order to deal with complex datasets that result in a high computational overhead and resource consumption, a new model that uses microservices to implement the random forest algorithm for malware detection and analysis has been proposed in this paper. The latter part of this paper discusses the overall microservice model and how it can lead to improved results overtime when the size of dataset provided increases by thousands. In conclusion a detailed difference between the micro- services-based model and the non-microserviced model has been presented by use of the acquired results gained during the application phase, the comparison details the parameters such as ROC curve, precision recall curve, bootstrapping resampling for AUC & a detailed feature selection of dataset to showcase the difference in values between the microservices model dataset and baseline model dataset.

In recent years, the increase in healthcare data demands the adoption of fog computing for fast processing and analysis of large data volumes. Fog computing enables real-time data computations that serves closely to healthcare devices. This study proposes the analysis and computations of EEG healthcare data using Pareto optimization-based fog computing. The proposed 2DP-FHS technique selects a fog head among heterogeneous fog devices to manage the EEG data within the fog layer. Additionally, an alternative fog head is designated to ensure continuous operation within the fog layer. The study considers various scenarios and unbiased fog devices in the simulation. Further, it presents the delay analysis varying the number of EEG and fog devices. The results show the effectiveness of the proposed technique across all scenarios highlighting its real-time applicability for various healthcare centers requiring time-critical EEG applications.

This study utilizes geotagged microphone technology to analyze vehicular mobility patterns and noise pollution in Indian suburban regions. Audio data capturing traffic scenarios were meticulously calibrated using a Sound Pressure Level Meter (SPLM) and manual markers to identify five mobility types. A novel approach employing a modified Convolutional Neural Network (CNN) was developed to accurately extract vehicles and characterize mobility patterns, achieving an impressive 92.1% accuracy rate. Particularly it shows significant performance in identifying light mobility events (95% accuracy) and honking events (65% accuracy), surpassing conventional method gaining knowledge of and deep learning strategies. Integrating the modified CNN into the pipeline superposed recognition technique by 4.12%. The version gives special insights into vehicular speed and noise characteristics, facilitating specific detection of mobility patterns. The advent of mobility maps contributes to Earth Science Informatics by using integrating geospatial analysis with advanced neural network strategies, imparting a comprehensive information of vehicular mobility and noise dynamics in Indian suburban environments.

Peritoneal carcinomatosis is a malignant cancer that spreads to the surface lining of a person's abdominal cavity and is usually caused by infection from other organs. AI developments, one of which is YOLO, can be used to help detect peritoneal carcinomatosis lesions. This research detects peritoneal carcinomatosis lesions by comparing several versions of YOLO with different scales, namely YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, YOLOv6sn, YOLOv6s, YOLOv6m, YOLOv6l, YOLOv8n, YOLOv8s, YOLOv8m, and YOLOv8l. Recall, precision and mean Average Precision (mAP) metrics are all used in this study as well as inference time. The results show that the recommended models are YOLOv8l and YOLOv5l where both get the same high results with mAP of 0.799, followed by YOLOv8s, with mAP results of 0.796. The study's findings are intended to direct future clinical applications and determine the most appropriate model for the identification of peritoneal carcinomatosis. This study provides in-depth information that forms the basis for informed decision-making, highlighting the accuracy required to address issues related to peritoneal carcinomatosis.

Health care is provided by health care professionals in the related fields of health and social care. The healthcare system provides healthcare services to ensure the health of the target groups. There is an increase in infectious diseases, and non-communicable diseases, which will create a double burden on the health of the public against it. It is best to try and find the right doctor and ask for a certain amount. It is not an easy task. This is for the health information management systems, helping you to find and match a specialist doctor for a particular illness in the immediate mode, and will also provide advice and information on the disease can be cured. This health management system makes it easy to find a doctor who will help the patients to recover more quickly from illnesses. Health care, in combination with information technology in a new direction to it is full, complete, and consistent information directly affects the quality and efficiency of the health service, as well as the safety, security, diagnostics, and treatment. However, the level of information is defined as a reflection of the degree of its application, and the degree of integration of the information systems, processes, and procedures. This concept can be defined as the integration of the information systems (is) and other applications are being used together, and can be accessed via a variety of communication networks. In order to fully carry out the integration of the IP, security, and a lot of medical information, as well as an effective, easy, and fast exchange, as well as for scientific analysis, and processing of tasks employing all available technologies—including cloud computing, Internet of Things, Internet technology, and data analytics—is essential. This research explains the pertinent principles and talks about earlier studies in this field. To integrate machine learning and other methods to examine the literature on health care management. Finally, the more accurate solution is provided in such scenario using machine learning algorithm, which makes it not a simple chat box, it’s a Integrated Model for the Prediction of Disease and Treatment Recommendation.

The integration of Artificial Intelligence (AI) technology has exerted a significant impact on the retail sector. Nevertheless, the adoption of AI carries substantial responsibilities and risks that fall squarely on the shoulders of senior managers. In this paper, the retail online dataset has been utilized. The contribution of the paper is an improved K-Means algorithm which has been customized for the segregation of product quantities by the top 5 customers. In addition to that, agglomerative clustering has been applied for the frequency of products which has been followed by the implementation of a Clustering algorithm to quantify individual customer purchases. The proposed DBSCAN algorithm categorizes products with similar costs by effectively grouping them. The outcome of the simulation results in the DBSCAN algorithm which exhibits improved performance in in terms of a substantial volume of data by sharing similar properties.

Machine learning models have emerged as the most accurate prediction choice in various fields in recent year. These include healthcare, agriculture, cybersecurity, finance, etc. Now a days medical science is one of the most benefited fields from machine-learning models. If the model is appropriately trained, machine-learning techniques can diagnose many diseases early. Early diagnosis and effective health management can save many lives from severe heart disease. This research paper proposes a new machine learning model, NeNBoost, for early detection of heart disease. It uses a National Government open source dataset based on the living style of the samples. To balance the dataset, SMOTE algorithm is used, and then an Artificial Neural Network (ANN) combined with Gradient Boost is applied using the features of the neural network. Compared to other ML algorithms, the proposed model’s efficacy is estimated based on its accuracy, precision, recall, F1-Score, and AUC-ROC score. The analysis signposts that the suggested model outperformed with the maximum accuracy of 96% in comparison to other models.

This paper explores the application of zero-shot learning, a novel machine learning approach, in predicting and mitigating cybersecurity threats and defenses. Our research bridges the gap between the theory and practice of using zero-shot learning, a technique that allows models to adapt to unseen scenarios, in anticipating and managing cybersecurity threats. We propose new attack methodologies based on zero-shot learning and discuss the development of robust defense mechanisms. An interdisciplinary approach, combining social engineering, behavioral psychology, and AI governance, underpins our work. Our research also introduces a quantitative model to assess the impact of these emerging threats and a proof-of-concept prototype. In closing, we discuss ethical considerations and advocate for responsible AI practices and robust regulations to prevent misuse. This paper underscores the potential of zero-shot learning to revolutionize cybersecurity practices and preparations for future threats.

Data engineering is a process involving extracting, transforming, and loading data to ensure that it is clean, reliable, and accessible in various analytical tasks. This research introduces an innovative approach to exploring nonverbal expressions in individuals with borderline personality disorder (BPD) using data engineering techniques. The study focuses on analyzing nonverbal cues exhibited by patients during a virtual tossing game to identify key factors influencing emotional responses in both inclusion and exclusion scenarios. By leveraging machine learning algorithms such as K-means, Decision Trees, and Random Forest, the research emphasizes the significance of head movement as a primary discriminator, followed by the task conditions and heart rate. Therefore, evidence of the effectiveness of using nonverbal expressions of patients with BPD by data engineering to offer featurization and new predictive values to support informed decisions of the clinicians is presented.

In Vietnam, rice stands as a vital food source for the whole population, and because of that timely identification and management of rice plant diseases is extremely crucial. Among the diseases that appear in Vietnam, blast leaf, leaf folder, and brown spot are among the most prevalent ailments that harm rice plants, directly impacting cultivation. To address this problem, deep learning, a cutting-edge solution can be used to detect plant diseases. In this paper, we present an approach utilizing Deep learning to identify illnesses affecting rice leaves. YOLOv8 is employed and trained with various strategies of Stochastic Gradient Descent with Warm Restart (SGDR) to identify diseases on rice leaves. The result of the proposed method is assessed using 309 images. The experimental findings reveal an accuracy of 88.6%, indicating superior performance compared to alternative methods.

The field of Speech Emotion Recognition (SER) has accumulated significant traction in recent years, driven by its potential applications in human-computer interaction, medical, and education. This study investigates the effectiveness of various SER models by conducting a comparative analysis using a specially curated dataset. The dataset comprises voice recordings from 129 female participants, each expressing seven distinct emotions while uttering the neutral sentence “The cat is sleeping”. This design gives SER models a richer and more realistic evaluation platform by allowing the collection of subtle emotional fluctuations within an impartial setting. Five well-known machine learning models are used and examined in this study. Each model’s performance is thoroughly assessed using important metrics including accuracy and F1 score, and the results are displayed using loss graphs and bar plots. A greater comprehension of the merits and demerits of each model in the context of SER is made possible by this thorough examination. The aim of this research is to enhance SER technology and its applications by comparing SER models, identifying research gaps, and suggesting future paths.

One of the key method in the machine learning process is regression analysis. There are several regression models available, and hence choosing the best regression model for a specific dataset might be challenging. This paper aims to conduct an experimental study for implementing eight regression models in context of machine learning. Eight regression models with various parameter values and regression measurement matrices are covered in this work. The experiment is conducted using several benchmarking datasets in order to determine the optimal regression model that is produced. Detailed comparative analysis is presented as the results. The experiment shows that the optimal regression model result depends more on the features selection and appropriate data imputing procedures, proper data cleaning than it does on the data type. The data cleaning process, which comprised appropriate methods for data imputation, feature scaling, feature extraction, and feature selection, determines the correctness of the regression model. Even though the data cleansing is done effectively, more research on data scaling strategies is necessary.

The localization and identification of advertisements play a pivotal role in content analysis and information retrieval. Acknowledging this significance, this paper focuses on the critical task of model evaluation and selection for robust and efficient advertisement detection. Employing a comprehensive methodology, we assess various deep learning models based on their accuracy, efficiency, and reliability in detecting and localizing advertisements within diverse print media formats. Our study reveals that certain models significantly outperform others in terms of mean Average Precision (mAP) and F1 scores, while also maintaining low inference latencies. These findings have profound implications for the development of more effective and efficient advertisement detection systems in print media. The conclusions drawn from our research provide valuable insights for future advancements in digital advertising and media analytics.

The human brain, among the most complex and mysterious aspects of the body, harbours vast potential for extensive exploration. Unravelling these enigmas, especially within neural perception and cognition, delves into the realm of neural decoding. Harnessing advancements in generative AI, particularly in the Image Processing domain, seeks to elucidate how the brain comprehends visual stimuli perceived by humans. The paper endeavours to reconstruct human-perceived visual stimuli using Functional Magnetic Resonance Imaging (fMRI). This fMRI data is then processed through pre-trained deep-learning models to recreate the stimuli. Introducing a new architecture named LatentNeuroNet, the aim is to achieve the utmost semantic fidelity in stimuli reconstruction. The approach employs a Latent Diffusion Model (LDM), emphasizing semantic accuracy and generating superior-quality outputs. Text conditioning within the LDM’s denoising process is handled by extracting text from the brain’s ventral visual cortex region. This extracted text undergoes processing through a Bootstrapping Language-Image Pre-training (BLIP) encoder before it is injected into the denoising process. In conclusion, a successful architecture is developed that reconstructs the visual stimuli perceived and finally, this research provides us with enough evidence to identify the most substantial regions of the brain responsible for perception.

The research introduces a proposal of indicators for the evaluation of educational applications aimed at children with autistic spectrum disorder (TEA). These indicators are classified as fundamental dimensions such as design, content and pedagogy, thus establishing an integral framework for evaluation. Through an instrumental analysis of the validity of content, the indicator system was validated, evaluating its clarity, coherence, relevance and objectivity through the expert judgment method. The results reveal a high interoperative reliability, backed by outstanding correlation coefficients. The validity of the system is reinforced with the unanimous approval of experts, who show a high degree of agreement in key aspects such as clarity, coherence, relevance and objectivity.

The positive results of the validation of the instrument have given rise to a system of indicators that cover essential criteria for the development of applications for people with ASD. Tea specialists and communication and information technologies (TCIS) now have guide elements for the design and development of applications adapted to the specific needs of these children, covering a variety of parameters.

Automatic Speaker Recognition (ASR) is indeed crucial for enriching transcripts by identifying individual speakers from audio files. ASR algorithms extract unique speech characteristics from audio signals, providing valuable information about speakers within a document. These algorithms rely on feature extraction techniques that should be robust to recording parameters while containing sufficient speaker identity information.

Machine learning plays a vital role in ASR, encompassing identification, verification, and detection of speakers. In a recent project, various machine learning algorithms such as k-nearest neighbors, SVM_SMO, MLP, Logistic function, and Naïve Bayes were applied to ASR, specifically used i-vectors for speaker recognition. These algorithms were evaluated on speaker identification and verification tasks across two different DataSet in English and Arabic languages. The evaluation criteria included the recognition rate (ASR Score/Accuracy), and the results highlighted the effectiveness of feature fusion for enhancing automatic speaker recognition performance.

The complexity of the digital landscape is on the rise. Thus, the urgent issue of personal data protection has become a major concern for individuals and businesses in the modern era. This article uses text mining techniques to examine data concentration and understand current prevalent threats. The data was collected from the “Have I Been Pwned” (HIBP) website, a globally recognized platform for tracking data breaches and compromised accounts. The data was then cleaned, tokenized, and input into a sentiment analysis tool in the Natural Language Toolkit (NLTK). Research findings highlight cybersecurity as a prominent concern, albeit one with a declining trend. Further scrutiny employing text mining methods reveals the existence of multiple issues. Concerns such as password intricacy and susceptibility to spam emails underscore users’ inadequate comprehension of security matters. Besides the objective reasons for purposeful cyberattacks, the consequences warn of risks related to carelessness and limited awareness of cybersecurity issues by users and data owners.

For countries with the large rice export in the world, taking care of rice plants for them to grow healthily is very necessary. One of the techniques is to detect diseases promptly, at the right time, on the right diseases in order that necessary treatment can be taken to minimize the harmful effects of diseases on rice plants as well as prevent them from breaking out into epidemics. Previously, disease detection was carried out manually, which took a lot of time and labor. Nowadays, automated methods have been increasingly applied for precision agriculture applications. In the article, a method to automatically detect rice diseases based on machine learning, especially deep learning concentrated method is proposed. In this article, our study concentrates on detecting four most common rice diseases: brown spot, hispa, blast and blight. The dataset for training, testing and validation collected and constructed from Kaggle and Google dataset includes approximately 400 images for each type of rice leaf diseases. From the experiment to assess the performance of the proposed learning method, the accuracy can reach approximately 97%. Thereby, the method shows feasibility and reasonableness, with extensively potential usage for practical precision agriculture application.

To reduce duplicated information system development projects, Korean government established the pre-consultation body. The pre-consultation body performs preliminary review for various aspects of government-managed information systems. In spite of its good roles and its effectively, the pre-consultation body met over-loaded duties, and now it shows some warning signals mainly due to the increased number of submitted projects. This paper statistically analyzes the results of pre-consultation processes for 1,399 submitted projects in the field of e-Government projects in the year of 2022. Our analysis shows that it is better to exclude some small-size projects from the pre-consultation process. This paper aims to provide the theoretical and statistics-based backgrounds to the claims, and we expect that our suggestions can improve the pre-consultation process of Korean government information systems management.

With increased competition and advancements in technology, Higher Education Institutions (HEIs) are forced to improve their communication and enhance their presence on social media. The purpose of the study is to examine and compare the communication strategies of HEIs on Facebook and Twitter in the context of Abu Dhabi, UAE. The study analyzed 4,148 Facebook posts and 6558 tweets from September 1, 2019, to August 31, 2020. This period was considered, covering both pre-Covid and Covid period. The comparison of posts on Facebook and Twitter revealed similarities and differences in terms of posting days, hours, and months. Both platforms heavily relied on visual content, particularly photos. Language choices varied among the platforms.

In the current era, major technology is cloud computing and Internet of Things due to their vast applications in day-to-day life. The major reason behind today’s internet traffic is big data transmission and usage in varied data types like textual, image and video datasets, as well as audio data. Due to widely spread usage of social media, security and authenticity of image data is of utmost priority especially when it comes to applications like medical images, Geographic imaging system, biometric authenticity. The major objective of this research is to provide a sustainable robust encryption algorithm for securing images where data authenticity is of priority like biometric images. Algorithm is lightweight cryptographic algorithm as it uses low dimensional chaotic map Arnold’s CAT map (ACM) and \(\mathscr {L}\)-System fractal for key generation. The results tabulated show satisfactory output towards statistical tests. Robustness and effectiveness of the proposed algorithm is compared decently with other existing algorithms and it demonstrates practical usability of the algorithm.

Efficiently detecting faults in steel plates is essential for maintaining the safety and dependability of structures and industrial machinery. Timely identification of faults mitigates further damage and averts exorbitant repair costs. This study delves into the efficacy of employing ensemble machine-learning classifiers for fault detection in steel plate manufacturing processes. Specifically, five powerful machine learning models—Random Forest (RF), AdaBoost, Decision Tree, Support Vector Machines (SVM) and Naive Bayes —are investigated in this study. The ensemble models (i.e., RF and AdaBoost) harness the collective power of multiple weak learners to enhance discrimination capacity. Evaluation is conducted using a publicly available dataset comprising seven distinct fault types: Pastry, Z_Scratch, K_Scratch, Stains, Dirtiness, Bumps, and Other_Faults. Results demonstrate Random Forest achieving the highest AUC of 0.942, with an accuracy of 0.771 and balanced F1 score, compared to the other models. This comprehensive investigation enhances fault detection efficacy, fostering informed decision-making in steel plate manufacturing processes.

The study and treatment of human infection continues to be examined in relation to generative AI, deep learning, machine learning, and AI (artificial intelligence). We provide an overview of AI’s current and prospective uses, as well as how it relates to clinical infection control. Research studies, Clinical trial, and meta-analytic received precedence when screening 1617 the PubMed database findings. The review’s narrative format centres on investigations that employ clinically validated prospectively acquired data from the real world, as well as studies with transformative possibility, including unique pharmaceutical discovery and microbiome-based interventions. Clinical imaging analysis (e.g., tuberculosis of the lungs diagnosis), tools for clinical decision-support (e.g., antimicrobial recommending, sepsis prediction), digital culture plate reading, antimicrobial resistance profiling, and malaria diagnosis are a few areas where there is proof to support the medical value of artificial intelligence (AI) used for diagnostics in laboratories. To date, most studies have not included medical metrics or real-world validation. Comparability is hampered by substantial variation in research methodology and reports. There are a lot of practical and ethical concerns, such as bias risk and algorithm transparency. Though the practical medical value of artificial intelligence (AI) tools for infections investigation and control seems to be much more modest, enthusiasm for the research and creation of these tools is certainly gaining momentum.