Proceedings of the 6th International Conference on Advance Computing and Intelligent Engineering

Mel frequency cepstral coefficients are one of the most prominent sets of primary features of an audio signal which are used for speech detection and cough analysis. This paper presents a new method that can overcome some of the common problems faced by using MFCCs for cough detection. In the proposed method, the most prominent part of the cough sample (HCP) is extracted and used to obtain the MFCC vectors of that particular window. These HCP MFCC vectors work as a standard comparison index for all cough samples to detect any respiratory disorders. The evaluation of the proposed method is done using 40 samples of COVID-19 patients of which 20 are positive and 20 are negative. The accuracy of the proposed method is compared with that of the standard MFCC method for the same set of samples. The proposed HCP MFCC method produces results that are 7.84% more accurate than the standard method. By bringing a standard set of comparing features that can work for almost all use cases, this method can be used as a quick identifying tool for various respiratory diseases.

Arrhythmia is one of the diseases that affects many people around the world. Deep learning provides an efficient tool to detect arrhythmia disease. A convolutional neural network (CNN) is an emerging technique used often for feature extraction in the medical domain. In this paper, AlexNet, VGG-16, VGG-19 models are used as the feature extraction method, and the selected feature is supplied as input to four well-known classifiers such as decision tree, kNN, LDA, and SVM for arrhythmia detection. Furthermore, an experiment is conducted with the combination of proposed CNN model where mRMR is used as feature selection method. Finally, the result of experiment is compared with different machine learning algorithms where LDA shows the efficiency in term of classification accuracy. The classification accuracy of the proposed model is recorded as 99.46%. The performance of the proposed model is higher in terms of classification accuracy compared to previous work on arrhythmia detection.

Epilepsy is a very common neurological disease that affects millions of people worldwide. Though there is no definite cure for epilepsy, predicting the onset of epilepsy in the preictal period will ensure proper health care which is administered during the ictal period. The proposed epilepsy prediction model uses the time series deep learning classifiers called the echo state network (ESN) and the InceptionTime network. The surface and intracranial EEG data from the University of Bonn’s Time Series EEG dataset are used for training the proposed models. The experimental results prove that the proposed deep learning model achieves an accuracy of 94.6% using ESN and 100% using InceptionTime network.

COVID-19, the virus that has affected current living standards, has undergone mutation, and the second wave has caused a much more devastating situation in India. In such a scenario, the alert of a third wave by the authorities has alarmingly increased concern in the nation. After being declared as an international emergency, the development of its vaccine has been conducted by different countries. India among other countries is also pursuing to develop much more efficient variants of the vaccine. The situation still persists to be hostile and maintaining the current precaution measures and maximizing the distribution of the vaccines is the only solution in hand. A necessity arises for a user-friendly app to reduce social interaction while assisting in medical support. In this paper, we have proposed an android application named YUDH, which focuses on the overall service that an individual requires from booking test centers, vaccine slot notification to home sanitization. The user can book COVID-19 testing centers and can arrange sanitization service after recovery with the provision of place, date, and time. In addition to booking test centers, swab testing at the doorstep is also available. The user also gets regular notifications on COVID vaccine slot availability in accordance with CoWIN portal and users’ preferences. This deployment is aimed at the safety of the user and their privacy safeguard. The application also assists the government to maintain a database more efficiently.

It is not very safe to go to hospitals for regular check-ups as we are used to. Patients can get medical advice from the comfort of their homes from specialized medical professionals. Patient data can also be digitalized, so that it can be used at any hospital. Symptoms are taken as input, and our deep learning model will give the possible disease, which the person might be suffering from, and the doctor he/she needs to consult for further medication.

The coronavirus epidemic began in Wuhan and has already spread to practically every country on the planet. Conravirus has a big population in India, and people are becoming infected at an alarming rate. Machine learning algorithms have been utilized to find trends in the number of active cases owing to COIVD in India and the state of Odisha in this study. The data was gathered from the WHO and studied to see if there was a link between the number of current cases, those who died, and those who recovered. The model was entirely based on multiple regression, support vector machine, and random forest which fits as an effective tool for prediction and error reduction. Based on the dataset taken from March 16, 2020, to August 20, 2020, from the ICMR website, the mean absolute error (MAE) of SVM is less for Odisha and multiple linear regression is less for India. The multiple learner regression model is able to predict number of active cases properly as its R2 score value are 1 and 0.999 for Odisha and India, respectively. Machine leaning model helps us to find trends of effected cases accurately. The model is able to predict what extent the COVID cases will grow or fall in the next 30 days which enables us to be prepared in advance and take some preventive measures to fight against this deadly COVID virus. It is observed that features are positively correlated with each other.

Despite the recent global concern, healthcare specialists, doctors, and scientists around the globe are still looking for a breakthrough solution to help fight the COVID-19 outburst. By use of artificial intelligence (AI) and machine learning (ML) in past, outbreaks have intrigued scientists, suggesting a particular methodology to tackling the existing coronavirus pandemic. In terms of the outbreak that followed after coronavirus, widely recognized as SARS-CoV-2. This paper provides an in-depth analysis of appraisal of AI and ML as one good approach for monitoring for contact tracking, prediction, forecasting, and therapeutic development.

Parkinson’s disease is a progressive neurodegenerative illness that causes movement abnormalities. Because of their dread of retropulsion, many with Parkinson’s disease refuse to leave their rooms and remain immobile. Injury-related permanent impairment exacerbates the issue. Deep brain stimulation is now the sole therapy option for the illness, but it is not accessible for everyone because it is more expensive, intrusive, and requires the installation of electrodes and a pacemaker. While existing methods fail to give long-term relief at a high cost, our discovery helps to slow the course of Parkinson’s disease non-invasively and also provides better therapy to the majority of the senior population with motor problems at a lower cost. Our concept is to create a wearable head cap with motors and drivers that would provide mechanical stimulation in the manner of the ancient Varma medical technique. As it has Bluetooth interference it can be easily connected to android and make it work accordingly. It can give care at home, making therapy outside of hospitals more convenient. We believe that our project’s originality and creativity will help us reach our aim.

As one of the main early adopters of innovative advances, the medical care industry has delighted in much accomplishment accordingly. In various wellbeing-related fields like new operations, patient information the executives, and ongoing infection treatment, artificial insight subset machine learning is assuming a key part. Inside the medical services industry, machine learning is gradually acquiring a foothold. An assortment of medical services circumstances is now being affected by machine learning (ML). With machine learning (ML) applied to the medical services industry, a great many different datasets can be investigated to make expectations about results, just as given opportune danger scores and exact asset designation. This exploration prompted the making of a more effective choice organization for clinical applications.

With the increase of micro-service-based architecture, detection of anomalies in enormous, complex production infrastructure has become complicated. In this work, authors provide a solution for finding anomalies using the structured event objects in a complex micro-service-based productions environment, where the flow of data is assumed to be periodic, deterministic and predictable in nature. As these objects are multivariate and multidimensional in nature, the number of features and dimensions is reduced without losing the quality of data. Next, a method to find anomalies has been proposed with the obtained dataset. The proposed method uses an unsupervised anomaly detection model using Tax–Duin approach for one-class support vector machine (SVM) to classify outliers. The basic assumption of this model includes classifying the first occurrence of any event as anomaly by default. The experimental results obtained indicate an accuracy of 88% by applying one-class SVM on the considered dataset.

The inceptions of generative adversarial networks have made it possible for machines to mimic creativity, one of the most unique and sophisticated human characteristics. Due to the rapid advancements in the field of generative adversarial models, lots of approaches have been proposed in the past. One of the most efficient GANs is deep convolutional generative adversarial network (DCGAN), which uses convolutional layers in the generator model to generate more realistic fake images. In this paper, we propose a lightweight implementation of the DCGAN that can be productive for the animation industry. Our model can be used by animators and designers to innovate ideas about creative anime avatars that have never existed before. This novel approach not only saves a lot of time on creative thinking but also provides brand new character designs for anime and manga avatars production.

One of the drawbacks of using predictive quality of service (QoS) in cloud service suggestions is that the values vary rapidly over time, which may result in end-users receiving inadequate services. As a result, the cloud-based recommendation system’s performance suffers. In this paper, an efficient service recommendation with a spatial–temporal aware QoS prediction mechanism in a cloud computing environment is proposed. The main contribution of this article is to use the geographical location of the services to help us choose the closest neighbor to show time QoS values sparingly, reducing the range of searches while increasing precision, and then using the Bayesian ridge regression technique to model QoS variations by making a zero-mean Laplace prior distribution assumption on the residuals of the QoS prediction, which corresponds to a Bayesian regression problem. The findings of the experiment show that the proposed approach may enhance the accuracy of time-aware cloud service recommendation by 10% over the previous approaches of temporal QoS prediction.

In today’s scenario where there is an exponential increase in the need for IoT services, user authentication is an essential feature for device security. IoT services allow many devices to be accessed and connected over the Internet anytime and anywhere and thus pose a risk over its privacy and security. A single-factor authentication scheme can be used to provide robust security, privacy, and secure access over devices. To deal with a range of IoT devices with varying storage, we have propounded a lightweight novel password-based user authentication scheme with low computation costs. This scheme uses lightweight XOR and hash operations. AVISPA tool has been used to prove security against various attacks.

For analysing networks like social media networks, wireless sensor networks, etc. in many applications, generating random connected graph is very important. As it is time consuming to generate the random connected graph consisting of large nodes it is necessary to generate it in minimum time. Characteristics like dependent edges and non-binomial degree distribution that are absent in many classical random graph models such as the Erdos-Renyi graph model can be captured by random graphs with a given degree range. The problem of random connected graph generation having a prescribed degree range has been addressed here. Random graphs are used to model wireless sensor networks (WSNs) or IoT comprising of sensor nodes with limited power resources. A fast and light-weight algorithm has been proposed in this paper to produce a random connected graph for a real-time multi-hop wireless sensor networks (WSNs). Results show that our method has better performance than other existing methods.

In a communication channel, there is a possibility of distortions such as ISI, CCI, and another source of noise that interfere with useful signals, and the signal becomes corrupted. Therefore, equalizers are needed to counter such types of distortions. In this paper, we presented a nature-inspired hybrid algorithm which is an amalgamation of PSO and GWO. The proposed algorithm is called HPSOGWO. During this work, we pertain to ANN trained with the proposed HPSOGWO in the channel equalization. The foremost initiative is to boost the flexibility of the variants of the proposed algorithm and the utilization of proper weight, topology, and transfer function of ANN in the channel equalization. The performance of the proposed equalizer can be evaluated by estimating MSE and BER by considering popular nonlinear channels and added with nonlinearities. Extensive simulations show the performance of our proposed equalizer, better than existing NN-based equalizers also as neuro-fuzzy equalizers.

In developing countries such as India, the transportation industry plays a critical role in the economy. Transportation is a vital means of transporting products and people from one location to another. Increased trade and business are facilitated by improved transportation, primarily railways. However, the railway is currently beset by several accident issues. The manual management of such a system is impossible every day. If it is done manually, it will take more time and money. A prototype for automatically examining and identifying cracks, obstacles, and fire in railway tracks is proposed in this paper keeping in mind the passenger safety and comfort. To save electricity, the planned system will turn lights and fans on and off automatically based on the number of people in the coach. The proposed model consists of an IR sensor, ultrasonic sensor, flame sensor, touch sensor, and an emergency switch. If a flame or obstruction is detected, the train will be delayed and eventually stop. The status of all sensors is updated on the Web site to assist railway administrators.

In this article, a new approach to modeling a nonlinear channel is proposed. In these works, a new cuckoo search algorithm trained with radial basis function neural networks (RBFNN) is applied in the non linear channel for equalization. The efficiency of the proposed algorithm enhanced through the search process by the integration of discovery and exploitation. Therefore, instead of Levy mutated step size operator in CS, the Cauchy mutated operator is used to create the step size which will make a random number which is used to generate a new solution for the global search. The performance of the proposed equalizer can be evaluated by estimating MSE and BER by considering popular nonlinear channels and added with nonlinearities. The consequences of the simulation show the presentation of our projected equalizer better than existing neural networks-based equalizers available in the literature.

Recommender system is one of the emerging niches of machine learning. They are being used widely in a variety of fields. Netflix, IMDb, and Amazon have been using it commercially for their interests. With the increase in usage of cloud services, cloud service recommendation has attained significant attention in recent years. Cloud service recommendation system aims at helping the user to find services he might prefer. This paper proposes a user-based collaborative filtering method to enhance it. Here, we take the preferences of an active user first and then try to compute users with similar preferences using a popular similarity measure like cosine similarity which computes the similarity scores by computing the cosine distance between users. Then, we recommend the services the active users did not invoke. Finally, we found the root-mean-square error with the various number of recommendations of our proposed approach thirty times and noted the average. We observed that the cosine similarity measure works better than a few other popular measures like Euclidean distance by 14.99% when the number of top recommendations is five.

This paper addresses the issue of movement epenthesis detection in isolated gestures. A novel method based on background subtraction and singular value decomposition to detect movement epenthesis in isolated gestures are proposed in this paper. The singular values of the absolute difference matrix obtained after the background subtraction provide a set of discriminative features to segment the movement epenthesis frames and sign frames. An adaptive threshold value is determined using the statistical properties of singular values for movement epenthesis detection. We tested our method on the ChaLearn LAP IsoGD dataset and the videos containing Indian Sign Language words. Experimental results show that our approach detected movement epenthesis frames with an accuracy of 91.14% and 93.73% on the ChaLearn LAP IsoGD dataset and the Indian Sign Language dataset, respectively.

Health issues in this pandemic situation were most challenging factor for the people where the only path was Internet and the trust worthy Web sites. This research uses a drug review dataset from UCI machine learning repository and filtered a part of dataset based on some psychiatric conditions and identified fourteen drug groups considering suffixes of drugs. By applying exploratory data analysis (EDA) and sentiment analysis (SA) on the drug groups, we have identified the best drug group as well as the less efficient drug group according to its rating and sentiment polarity. From the experimental results on the considered dataset, it was found that barbiturate is the best drug group by customer rating with mean rating of 9.625, and antipsychotics drug group is found to have less rating as per customer suggestion.

In recent years as the computation power and availability of the data has increased exponentially, there has been significant increase in study of human sentiment in various fields. This paper examines the use of sentiment analysis in algorithmic trading. Macroeconomic variables such as GDP, Internet consumption and various other socio-economic factors are also taken into consideration in this paper. The main aim of this paper is to determine all factors and technical indicators that would give us a proper analysis. Human sentiment affects human behaviour adroitly, and thus, market is also not acquitted from its effect. This survey presents current advances in natural language processing (NLP) and prerogative positions of algorithms in market.

Machine perception has steadily become more accurate with deep learning methodologies. Abundant multimedia data sources have made navigating audio data essential. This work performs environmental sound classification as a step toward integrating artificial intelligence in audio data. Audio files are converted to tensors, resampled and then converted to mel spectrograms to account for human sensitivity to different audio frequencies. Pre-trained and high-performing convolutional neural networks (CNNs) are leveraged to train the ResNet-152 and DenseNet-121 architectures for transfer learning. The custom ensemble model uses these models for inference. The outputs of the models are combined and passed through a dense layer to generate an ensemble capable of inferring correct weightage for each of the models without manual interference. The ensemble model achieves promising results with an accuracy of 91%, and precision and recall of 0.91 and 0.93, respectively. The results demonstrate that a CNN-based ensemble method is adept at extracting and generalizing temporal information from audio signals.

Product recommendation becomes to be one most revenue-generating technologies that every e-commerce website is using. For enhancing the purchase rate and user engagement, various product recommendations are available in the e-commerce website. Most of the external contexts are also taken into consideration for the product recommendation. This type of external context analysis sometimes will provide better recommendations when compared to the user-generated data. This paper proposes an idea of product recommendation using the weather. Here we predict the weather for the coming days using logistic regression and analyze the weather using big data analytics. Based on the analysis, we will sort the products and recommend a product in the same weather.

Globally, the use of cryptocurrencies to purchase goods and services has been rising. They rely on a secure distributed ledger data structure; mining is an integral part of such systems. The rise of cryptocurrencies’ value on the market and the growing popularity around the world open several challenges and concerns for business and industrial economics. Cryptocurrencies have been triggered by the substantial changes in their prices, claims that the market for cryptocurrencies is a bubble without any fundamental value and also concerns about evasion of regulatory and legal oversight. Machine learning is part of artificial intelligence that can make future forecastings based on previous experience. In this paper, methods have been proposed to construct machine learning algorithm-based models such as linear regression, K-nearest neighbour(KNN), and also statistical models like Auto-ARIMA and Facebook’s Prophet (Fbprophet). This paper presents a comparative performance of machine learning and statistical modelling algorithms for cryptocurrency forecasting.

Indian Sign Language (ISL) is used in the deaf community all over India. Development of the ISL recognition system is an active area to aid this community. In ISL, most of the signs are two-handed signs, and thus, it differs from another commonly used American Sign Language (ASL) and seems complex. In this paper, the design and implementation of a system to recognize ISL signs is reported. Building such a system can help specially abled person/people, by providing a medium to communicate with others without human interpreters. The proposed system is built using a deep convolutional neural network (CNN), which performs both feature extraction and classification, preceded by an image preprocessing step. A real-time input (live signs captured from webcam) is given to this system, and the output is delivered in the form of text and audio. Proposed CNN architecture has achieved an accuracy of 98% for a given dataset which comprises 56 items (1–10 digits, A-Z letters, and 20 general words).

Facial analysis is an active research topic in examining the emotional state of humans over the past few decades. It is still a challenging task in computer vision due to its high intra-class variation, head pose, suitable environment conditions like lighting and illumination factors in behaviour prediction and recommendation systems. This paper proposes a novel facial emotion representation approach based on dense descriptors for recognizing facial dynamics on image sequences. Initially, the face is detected using the Haar cascade classifer to extract the temporal information from the facial frame by applying a scale invariant feature transform by combining a bag of visual words. Later, the extracted high-level features are fed to machine learning algorithms to classify the seven emotions from the MUG dataset. The proposed dense SIFT clustering performance was evaluated on four different machine learning algorithms and achieved a high rate of recognition accuracy in all classes. In the experimental results, K-NN exhibits the proposed architecture’s effectiveness with an accuracy rate of 91.8% for the MUG dataset, 89% for SVM, 87.6% for Naive Bayes, and 85.7% decision tree, respectively.

Unconstrained handwritten identification is among the toughest situations in recognition and image processing. This appears to be a simple operation for an individual, however, acknowledging handwriting is a time-consuming effort for a system. In the context of a device, the entry should first be digitized from a record, a photograph, or a legitimate device such as a desktop, tablet, or laptop. The digitized text or numeral is then changed into digital form text using the Handwritten Character Identification method. This could be managed and done in two ways: online plus offline. The central target of this survey is offline authentication of Gujarati scripts (characters plus numerals) in paper and electronic materials. Numerous neural and machine learning frameworks with classification techniques were being utilized, however, the bulk of machine methodologies demonstrated efficacy in spotting these scripts in the end.

With the introduction of web 2.0, individuals now have a perfect platform to communicate their ideas, thoughts, and feelings. Web Opinion Mining / Sentiment Analysis is a text mining job aimed at developing a system that automatically extracts, recognizes, and categorizes user opinions from natural language text, user provided content, or user generated media. We have employed classifiers to analyse data sets from a wide variety of domains. The ensemble algorithm is applied to boost its performance. With an delicacy of 80.93%, the Logical Regression using Ensemble Classifier exceeds the others.

This paper attempts to examine the performance of preprocessing strategies with logistic regression classifier. The goal of this paper is to see if there is a feasible and efficient strategy to enhance the performance of classification techniques on imbalanced datasets for different training dataset percentages. The experiments were conducted on Cleveland dataset—binary class. Several data preprocessing methods like Smote, Borderline-Smote, and ADAYSN were applied to data in order to classify various training dataset percentages. It was necessary to ascertain how the training dataset percentage affected the final classification for preprocessing methods. The experimental results explained that the ratio of 70–30 datasets performed better or better than other ratios when on train and test datasets, respectively. It was found from experimental results that the algorithms gave better accuracy when the training to testing ratio was 70:30 compared to other ratios.

Chatbots is the current trending topic in machine learning. It processes the user's queries and gives appropriate response. Chatbots are kind of virtual assistant where we talk with a computer bot not with a real human being. But it feels like we are taking with a real human being. Chatbots are being used heavily in various sectors like in banking, ticket reservations, customer enquiry desks, etc. Chatbots are helping businesses to give quick responses to the user queries, and most importantly, it is saving both time and resources of the businesses and customer. In future, chatbot will play major role for the communication between business and consumers for sure. Most of the chatbots are developed in English language by using various frameworks. This paper explains various techniques and architectures which can be used for developing a real-time chatbot.

The rapid development of image editing applications has brought about an immense number of doctored photos coursing in our day by day lives, encouraging an exorbitant interest for automated forgery detection algorithms that can rapidly check the legitimacy of the source image. A robust forgery detection system needs to process the images without the need for prior knowledge about the image or any integral watermark. Forgery detection systems can be developed based on several indicators extracted using the image production approach and by analysing the doctored photos for any abnormal behaviours when compared to the original image. Digital images are considered to portray visual information in the current era and have become quiet predominant in our day-to-day life. Digital images are predominantly used in several sectors for analysing purpose. Few of these sectors include clinical science, criminal assessment, etc., where the legitimacy of an image plays a crucial role. Prediction of forgery in digital images is considered to be the most crucial part of any investigation. In the subject of digital image fraud identification, adequate quality work has depleted the first decade. Detection of forgeries in a digital image and the application standard tools seems to be more optimal with respect to time, space, and ease of work. The methodology proposed over here applies a hybrid combination of deep learning (DL) and machine learning. The proposed hybrid combinations segregate the image as forged and non-forged images. The proposed approach also explores the application and effectiveness of neural network to predict forgery images.

Background: In recent years, speech recognition technology has become a dominant part of our everyday lives, and as most of the future technology being developed can easily be integrated with the help of speech recognition. To make a digital future, technological advancement of everyone is necessary, and to make this technological advancement not so technical, speech recognition serves its role. Although speech recognition has made significant advances at certain languages, what has been achieved is a drop and what is left is an ocean. This technology has failed miserably in recognizing different accents of a single language or a voice disorder, and this has led to various questions on the authenticity of progress of the process. This paper documents the drawbacks of this technology and the areas where its immediate progress is possible. It talks about the limitations of various existing and popular and under radar ASR technologies with insights of their flaws which need to be considered immediately to avoid various social dilemmas and insecurities.

Gesture recognition task is a challenging research topic due to different hand sizes and poses. A hand gesture recognition convolutional residual network that learns to partition the hand region from the background is proposed. The second stage employs two convolutional neural networks which takes appearance information and the shape information as input to perform the gesture recognition. The appearance information is given by RGB images, and shape information is given by segmented gesture region obtained from first stage. The classification is computed based on accuracy, precision, and F1-score. The results manifest a high accuracy of 98.75%, F-score of 0.94, recall of 0.94, and a precision of 0.95 which is better than other methods on OUHANDS dataset.

Two techniques of machine learning are supervised learning and unsupervised learning. In Supervised learning, data is labelled whereas in a non-supervised learning algorithm input data is unlabelled. Supervised learning is further divided into classification and regression and unsupervised into clustering. This paper discusses the idea of clustering and its classification into hierarchical and partitional clustering, further discussing the types of partitional clustering, mainly K-means, and its difference over partition around medoids (PAM) and clustering large applications (CLARA) is also explained. This paper talks about the application of K-means clustering in image compression, and a practical case of compressing an image is also discussed.

Segmentation of image is the method of separating objects from its background. It is helpful finding and deciding about which pixel belong to which objects. It is the process of making pixels of every region has similar visual characteristics. The watershed algorithm provides a complementary approach to the segmentation of an object. It is essential for segmenting objects where they touch their boundaries. Watershed is a dividing ridge between drainage areas. In digital image processing, the banks are the watershed lines and the drainage areas are catchment basins. Watershed is the representation of the grayscale image as topographic relief. Watersheds with adjacent catchment basins are being built during the repeated flooding of the gray value relief. This flooding method is performed on the gradient image. The magnitude of gradient value is intensely sensitive to image noise. So noise value plays a vital role in watershed processing. In computer-world watershed is a classical algorithm used for image segmentation. In this paper, we will witness the approaches of the watershed transformation in an image with a proper analysis of its advantages over other methods.

Human emotion recognition is an active research topic in analysing the emotional state of humans over the past few decades. It is still a challenging task in artificial intelligence and human–computer interaction due to its high intra-class variation. Facial emotion analysis achieved more appreciation in academic and commercial potential challenges mainly in the field of behaviour prediction and recommendation systems. This paper proposes a novel scattering approach for recognizing facial dynamics using image sequences. Initially, we extract the temporal information from the facial frame by applying a saliency map and hyper-complex Fourier transform (HFT). Later the extracted high-level features are fed to the scattering transform method and machine learning algorithms to classify the seven emotions from the MUG dataset. The performance of proposed wavelet scattering network was evaluated on four different machine learning algorithms and achieved a high rate of recognition accuracy in all classes. In the experimental results, K-NN exhibits the proposed architecture’s effectiveness with an accuracy rate of 97% for the MUG dataset, 95.7% for SVM, 93.7% for decision tree and 91.2% naive Bayes, respectively.

Day-by-day, a huge volume of data are generated through electronic gadgets via recent technologies. The data are generated in different forms and thereby its complex in nature. Therefore, big data analytics techniques are introduced to analyse the complex dataset efficiently in different perspectives. Among the different perspectives, the finding of frequent pattern matching in the given dataset is addressed in this work. However, researchers introduced various algorithms for finding frequent pattern matching. But, the existing algorithms have less accuracy for predicting frequent pattern matching and take more retrieval time. In addition to that, the existing algorithms do not preprocess the dataset. Therefore, the main objective of the proposed work is to increase the accuracy by preprocessing the dataset and minimize the response time by parallel processing the dataset. In order to achieve the above objectives, this work introduces an integrated frequent pattern (IFP) growth algorithm to find the distributed frequent pattern effectively from the large dataset. Therefore, IFP growth algorithm deploys in a Hadoop platform. In Hadoop, the Hadoop distributed file system (HDFS) processes the dataset using a multiprocessor. Henceforth, the IFP growth algorithm improves the accuracy and also minimizes the prediction time. Thus, the proposed work chooses a supermarket dataset as a case study.

Agriculture is crucial for human survival and is a major economic engine across the world, particularly in emerging countries. Plant seed classification is a multi-class dataset with 5,539 pictures divided into 12 classes. We investigate various learning classifiers for the image-based multi-class problem in this study. We will start with a simple convolutional neural network (CNN) classifier model and work our way up to more complex options like support vector machines, and K-nearest neighbors. We will create an ensemble of classifiers to increase the current state-of-the-art accuracy. We will also investigate data preprocessing techniques like segmentation, masking, and feature engineering for an improvement in the overall precision. We will compare the performance as well as their impact on the final ensemble. To overcome this challenge, traditional techniques use complex convolution layer-based neural network architectures like Resnet and VGG-19. Though these techniques are effective, there is still scope for increasing accuracy. In this study, we propose a boosting ensemble-based strategy that employs a multilayer CNN model with a deep convolution layer that is boosted using the K-nearest neighbors lazily supervised learning technique. Although the fact that this combination is less complex than previous ways, it has obtained a higher accuracy of 99.90%.

Text processing in document images has evolved as the most prominent research area these days attracting a lot of interest from researchers across the globe. It encompasses a range of activities from simple text detection to text translation. However, there are a good number of languages for which very little or no research work has been done. Most of these languages lack basic resources to carry out research. Kashmiri language based on Perso-Arabic script is one among such languages, which has been discussed in this paper. This paper presents a brief introduction of text processing in the image domain, a description of the Kashmiri language and related issues for kashmiri text processing in the image domain.

Face Detection is a famous topic in computer vision. Over the most few couple of years researchers have attempted to improve the performance of face detection algorithm in plane and out of plane rotation. In this paper, we propose a quick way to deal with face detection algorithm using support vector machine (SVM) and golden ratio. For performing this new algorithm, the main prerequisite is the preparation dataset in the front facing appearances to prepare SVM for skin filtering. In the proposed algorithm first we apply color histogram equalization (If the face detection algorithm is not able to detect any face) which can address the mistake of the skin filter then apply SVM for removing non-skin color, i.e., a skin filter machine is developed using SVM and lastly apply golden ratio for detecting the face region correctly. Proposed algorithm is compared on three datasets XM2VTS, FERET, and BioID with a high discovery rate not less than 95%. The experimental result shows the proposed algorithm not only runs comparatively fast but also gives an upgrade performance.

The Earthquake is an essential problem in human life, by using machine learning techniques in earthquake prediction, we can save humankind. Using the successful application of machine learning techniques indicates that it would be possible to use them to make accurate forecasts to avoid short-term earthquake damage. In this paper, with the first aim, we have applied seven machine learning techniques, namely, Artificial Neural Network (ANN), Decision Tree (DT), Logistic Regression, Random Forest Classification, Naïve Bayes (NB), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) to reach the best technique for prediction. The second aim used the methods of two learning styles, surface and deep learning, in training students with programming skills to use the seven techniques. Through two experimental groups, one of them used the method of surface learning (collective), and the other used the method of deep learning (individual).This is to determine the best learning style to teach students programming skills.

The paper aims at presenting Internet of Things in global business relating to sustainable development with the emphasis on its prospects and problems. Secondary data have been used in conducting the paper. Internet of Things presently is at an infancy stage with some devices which emerges with the revolutionary change in information and telecommunication systems that facilitate the growth of trade and multinational companies during globalization. At this juncture, cloud computing starts using as one of the important transformations of information technology, where computer along with Internet provides many facilities with a view to achieving goals of both companies and end users. Production systems starting from 3D printing and high-performance computing to the Internet of Things, and industrial robots are facilitated through cloud computing. During the study, it is clarified that IOT can be applied in many areas as wearables, health relating activities, traffic monitoring, fleet management, agriculture, hospitability, smart grid and energy saving, water supply, and maintenance management. The study also points out some limitations of using IOT in case of control and reliability, compatibility and contacts, and lock-ins. Despite the limitations of using IOT, it has emerged as a boon of human civilization at the age of globalization. With the passes of time, pattern of business and its management style is changing fast. For sustainable development of a country in particular and the world in general, learning and implementation of IOT may be considered to be an alternative technique in the world.

The worldwide focus on education has led universities to opt for online learning since the COVID-19 pandemic. Although online learning is becoming the new normal, there is minimal research on learners’ attitudes and behaviours towards online learning. The purpose of this paper is to explore the current behaviours, attitudes and learning experiences of students towards online learning through both students and academics perspectives. A mixed approach method was adopted, with a focus group discussion conducted via Zoom videoconferencing with eight academics and a survey-based questionnaire of 520 students via Google platform from the Open University of Mauritius. The main findings of this paper showed that learners need a sense of community between learners and instructors whereby there was a feeling of isolation. Hence, they value online discussions and opine that learning activities promote interaction with others. Making sure to study on a regular basis followed by attending online classes were the top characteristics of the learners. As for the academic perspective, low attendance was a major issue while rewards and teamwork were salient characteristics that came out from the findings. The paper concludes with limitations and suggestions for future work.

Organisations are eager to use and develop decisions by managing data in the most efficient way. To become more competitive and better meet their current and future needs, data-rich information enables managers to prospect the market, use the information for analysis and make decisions that could keep their businesses one step ahead of the rest. The scarcity of such data creates gaps in research information for business intelligence. Although algorithms and artificial intelligence are introductory concepts of data-driven decision management, the dashboard can be a useful tool that organisations can use both to present useful data and to make good decisions effectively. This paper describes the concept of using and managing data for decision-making in areas other than information technology. Using a conceptual framework to describe the importance of data-driven decision management, it argues that the creation of dashboards could be applied effectively the Université des Mascareignes (UdM) where it could meet the needs of users and decision-makers willing to get a clear picture of the problem affecting them. Rather than just disseminating information, the dashboard could provide important information that improves decisions and actions that can be initiated more efficiently and productively.

Classroom scenarios have long shifted from their traditionality in the modern era of blended pedagogical strategies to accommodate the didactic reality of the mixed abilities setup. The shift from offline to online mode has undeniably necessitated a thorough examination of the curriculum and methodologies adapted to ensure effective teaching–learning through the introduction of interactive digital tools, platforms, and resources. The emergence of digital drama, which includes humanities and technology-based tools, in tertiary classroom settings establishes interactions between lecturers and “net generation” learners. Our objective is an attempt to cope with the challenges and constructive potentials to enhance their French language skills with drama as well as foster empathy and socially competent behaviours. In adopting a mixed research method, the paper aims at identifying and scrutinising the social skills developed, acquired and adopted by the sample student population at the Université des Mascareignes, Mauritius.

Conglomerates in Mauritius are leveraging social media as a marketing tool. However, studies on the implementation of social media marketing by companies are limited. This paper aims to explore the internal organization of businesses in Mauritius to use social media as a strategic marketing tool. A qualitative research design has been developed to gather data from Chief Communication Officers and Chief Marketing Officers of major conglomerates in Mauritius. Semi-structured interviews were conducted. This study shows that conglomerates either manage their social media marketing in-house or they hire an agency to manage their social media accounts. To guide behaviors of employees, all conglomerates have developed a social media policy. Social media marketing requires specific skills and competencies and has given rise to new job descriptions. This study adds to the body of knowledge of social media marketing by providing insights about the approaches conglomerates have adopted to leverage social media.

Distance learning is currently of national interest in this period of confinement following the Covid-19 pandemic. In Mauritius, for example, the decision to suspend classes by the government is to ensure pedagogical continuity with online learning and teaching on the platforms of the Université des Mascareignes or via other platforms of e-learning. This article examines the measures taken by the authorities to ensure the quality of learning in terms of educational and didactic effectiveness and efficiency. This prompts us to take a new look at the issue of distance learning (DL), from the point of view of educational policies, and on the side of innovative approaches and emerging practices such as hybrid learning.

With the evolution of the Internet of Things (IoT), marketing has moved from a product-centric marketing approach (Marketing 1.0) to a customer-centric era (Marketing 2.0) further to a period where value-driven approaches were prioritized (Marketing 3.0) to ultimately focus on Marketing 4.0. This new era of marketing has seen customers to be more participative, more curious of something that is new and different, and businesses are focusing on machine learning (ML) to understand customer’s psychology. Traditionally, marketing was considered to be the synthesis of the four main elements of the marketing mix to satisfy customer needs, and there was a one-way communication: business to consumers (B2C). However, two decades after the twenty-first century, consumer experiences, information management, and predictive analytics are main factors to determine future trends. Practitioners argue that Marketing 3.0 is similar to Marketing 2.0 where customers’ involvement and participation contribute to the overall business marketing process, but what is new is emotional marketing where individuals’ needs and aspirations are also considered. With the advancement in information communications technology (ICT) and Web 3.0, the emergence of social media platforms has transformed the communication channels in marketing. Social media is an important marketing communication tool where customers share their experiences among themselves and businesses as well. Consumers’ shared experiences, information, and knowledge management together with predictive analytics are the main factors that are used in Marketing 4.0 to predict future consumer buying behavior. The aim of this paper is to go beyond the acceptance of ICT and understand its contribution in the future of marketing.

The change-point detection method based on decision-making statistics is considered. The average number of delay steps of the change-point detection and the average number of delay steps of detecting the returns of the statistical properties to the initial values are constructed as the mathematical models for the algorithm based on decision-making statistics with the signals ratio equations. The mathematical models for the average number of delay steps of detecting the returns of the statistical properties to the initial values are proposed for the first time. Proposed mathematical models for the average number of delay steps of the change-point detection have no limitation on the values of the algorithm's parameters, unlike already known ones. To ensure that the change-point detection in real-time systems will be done with a delay not exceeding a given value, it is critical that the obtained models already allow choosing the algorithm's parameters with the worst-case orientation.

An efficient procedure to obtain information on lighting levels, uniformity, energy consumption, and energy classes regarding street lighting through GIS is proposed in this paper. It shows the illuminated street along with its plot in terms of map data. The proposed methodology captures a nocturnal image of an illuminated street and subsequently, it determines the values of the average illuminance and the electrical power consumption on each street. The map is explored into a GIS which can provide the amount of power consumption installed along with corresponding energy classes from the street’s length. This proposed work highlights on installation of illuminance of the street lights as much as needed to cover the pavement area with least cost over the existing work.

This paper introduces a new hybrid approach that uses the output of Taguchi design of experiment (DOE) matrix to train an adaptive neuro-fuzzy inference system (ANFIS) model, and the rules of the ANFIS model are used to perform multiobjective optimization. The proposed approach is applied to optimize the geometry parameters of a flat plate heat sink such as the heat sink is length, width, fin height, base height, fin thickness, and number of fins and the multiobjective functions are minimization of the thermal resistance and emitted radiations of the heat sink. Also, the trained ANFIS model is used to predict the performance of the multiple outputs given the combination of input parameters. The multiple responses such as emitted radiations and thermal resistance are optimized by 22.18% with respect to the original design. Also, this method serves to reduce the search space for a given problem. Using the reduced search space, genetic algorithm (GA) was used at the second level of optimization to further improve the performance by upto 34.12% compared to the original design. Thus, the proposed Taguchi-based ANFIS modelling can either be used as a simple and standalone approach for optimization or can be combined with GA in the next level for any kind of MIMO parameter optimization problems.

A new fuzzy energy function for fuzzy quadratic programming problems is constructed using fuzzy norm. Based on the fuzzy energy function, a new fuzzy neural network is developed for solving fuzzy quadratic programming problems numerically in which all or some parameters are fuzzy. The stability of the proposed fuzzy neural network is established, and numerical examples are demonstrated to substantiate the significance of the proposed fuzzy neural network.

The loadability of a large power system network is investigated in this paper, and a particle swarm optimization (PSO)-based technique is proposed to evaluate the optimum position and setup of the FACTS device to optimize loading margin and voltage stability. Voltage stability is one of the most problematic aspects of power system functioning. There are numerous methods for determining the voltage stability of a system. The calculation of the margin from the present point of operation to the MLP is one of these approaches. The maximum load capacity is determined using this approach. It is to increase the voltage stability of power network in critical operating circumstances; it is recommended for STATCOM usage. The search for the optimal solution planning is expressed as an objective function toward the least real-power loss in the particle swarm optimization technique. An optimization challenge may be seen as the difficulty of identifying the maximum load point. The proposed system’s performance is assessed in a range of operational settings, including without STATCOM and with PSO-optimized STATCOM.

Online shopping has made the hassles of stepping out of homes to purchase items a thing of the past. With various organizations running their businesses online and with the onset of the pandemic, relying on e-commerce sites has simplified our lives. Although these applications provide details of products through the use of images and text, consumers are still wary and skeptical about the quality or fit of a product before making the purchase. In this work, we have discussed the use of virtual reality (VR) in these applications, to create a shopping environment for the consumers from the comfort of their homes. We have also proposed an architecture framework to implement VR on e-commerce. We provide a use case of online shopping which involves a product such as apparels along with certain future research directions.

Finding the associations among the itemsets and discovering the unknown or unexpected behavior are the major tasks of rare pattern mining. The support measure has the main contribution during the discovery of low support patterns. As the association of low support patterns may generate a bundle of spurious patterns, other measures are used to find the correlation between the itemsets. A generalization of frequent pattern mining called periodic frequent pattern mining (PFPM) is emerged as a promising field, focusing on the occurrence behavior of frequent patterns. On the contrary, the shape of occurrence in the case of rare pattern mining is not much studied. In this paper, a single scan algorithm called $$ PRCPMiner$$ PRCPMiner is proposed to study the shape of occurrence of rare patterns. The proposed algorithm discovers periodic rare correlated patterns using different thresholds with respect to support, bond, and periodicity measures. The research shows the influence of these thresholds on the runtime performance for various datasets.

Currently, computers and the Internet are used to conduct the majority of business transactions, communications and the automated control of industrial equipment, among other things. Working online makes the process more efficient and convenient. The risk of cyber-attacks has also increased significantly as a result of devices being exposed to the Internet on a daily basis. The Internet’s speed, ease of use and invisibility, lack of geographical boundaries cyber financial crimes, stalking and bullying are becoming more commonplace, according to the FBI. A digital forensic investigation carried out with the assistance of software tools yields evidence against cybercriminals that can be presented in court. This review work aimed to evaluate and compare the performance and applications of ten online digital forensic tools. The conclusions, limitations of these tools and how after moral improvement, they can be used to assist digital forensics professionals in discovering digital evidence are presented.

Over the nighttime, a vehicle’s headlights cause significant risk. Its high beam that comes from the opposite vehicle creates a temporary glare to drivers. The light intensity from the high beam is 1200 lumens which are comparatively higher than the light intensity from the low beam (700 lumens). For this circumstance, the headlight should be dimmed to avoid road accidents. To avoid such incidents, the automatic headlight dimmer circuit can be used. This circuit consists of light-dependent resistor (LDR) sensors and a night vision camera to sense the light from the approaching vehicle that comes in opposite direction. Then, the light signals are processed using an Arduino microcontroller and the output signals are given to the switching relays. Hence, it switches automatically from the high beam into the low beam and reduces the glare effect faced by the drivers coming from the approaching vehicle. It avoids human intervention in the dimming of headlights.

LCL filter design is difficult; still know the figures and facts of it in power industry application. To the mitigation of distortion, reduce bulky inductor size and reduce grid harmonic current cause of high switching frequency in inverter by zero voltage switching. LCL filter is more attractive due to low inductor size over conventional inductor filter. Designed suitable LCL filter, enhanced power factor of grid due to low loss dissipation in damping resistor of filter with low EMI. In this paper, the design of filter with mathematical model to verify the design procedure is carried out. The performance of system is simulated under MATLAB/SIMULINK environment. Reduced total harmonic distortion is shown in the grid current by using various filters are observed in simulated results.

Data warehouses bring an organization’s data together in one place for reporting and analysis. Top-level management receives key performance metrics, middle-level management receives analytical strength and bottom-level management receives consistent and accurate data based on analytical system information. Despite the numerous domain-specific challenges, data warehouses must be financially viable to be successful. We conducted a comprehensive literature review and read case studies of organizations that have successfully implemented data warehousing solutions in order to compile a list of best practices that can serve as guidelines for new practitioners and organizations new to this technology.