Machine Learning in Information and Communication Technology

Ours is the era of cardiovascular disorders. In this work, a corpus of ECGs collected in the state of Jammu and Kashmir is studied for automated perceptual similarity-based disease tagging. Following on our earlier work, we have deployed Earth Mover’s distance (EMD) as the similarity metric to generate automated disease tags. Rationale for generating these tags is based on the similarity of test ECG with healthy and unhealthy ECGs. If the test ECG resembles representative healthy ECG, then it is tagged as healthy, and if it resembles representative unhealthy ECG, then it is tagged as unhealthy. Future directions for increasing the accuracy of this work are discussed. It’s integration with biomarkers in a multi-sensor data fusion-based automated CVD tagging criteria is also explored.

Heart disease has been a major threat that costs human lives. There are many reasons behind heart disease including smoking, heredity, and diabetes. Day to day, people face various common symptoms of heart disease which are unconsidered in a lethargic manner. This leads to serious and life-threatening complications. To predict these diseases in prior, several methods are existing which take in a certain number of parameters for prognosis. The system proposed here is an ensemble approach that combines the idea of the MT-MrSBC algorithm along with bagging and boosting. The algorithm mentioned here overcomes the issues faced by other algorithms in handling the multi-variant environment. The algorithm deploys iterative techniques indulging bagging and boosting concepts that enhance the system. The system trained is thus capable of predicting the disease of the patient. This helps in taking precautionary measures by the patient which are life saving.

Chakraborty, Tiyasa Bhadra, Ashok Kumar Nandi, DebashisIn real medical applications, proper measurement of cancer disease area is very important, particularly so, if we want to compare the disease region between pre-treatment and post-treatment CT images for the same patient. The segmentation of a specific region can be defined as the grouping of image pixels corresponding to specific features. Several supervised approaches are there to solve the region segmentation problem. In our problem, we want to find the cancer area from the pre-treatment and post-treatment CT images which are in the state of raw medical data. In this study, we are using the SVM to measure the cancer area from the raw images where new features are taken into the study including neighboring pixels’ influence. Finally, our proposed approach is compared with the state-of-the-art methods which subsequently proves that our method performs more efficiently than other concerned procedures.

K-means transformation, histogram equalization, linear contrast stretching, and share-based features are all used to detect leukemia. A method for automatically classifying leukocytes using microscopic images is proposed. This proposed model used MATLAB to find leukemia cells in healthy blood cells, and it requires no medical equipment or expert and heavily relies on automation. This technology can detect anemia, malaria, vitamin B12 deficiency, and brain tumors. The proposed method correctly identifies WBCs and leukoblasts in images and refines the identification, thresholding, and segmentation phases. This improves WBC counting and overall segmentation accuracy, which leads to better shape feature extraction, which is critical for this problem. New features for this type of analysis must also be studied and analyzed. Finding the most discriminatory features will provide the best accuracy. Determining whether adjacent leukocytes can be separated is critical for counting all leukocytes in an image.

Leukaemia is one of the blood malignancies because of the strange expansion of white platelets in the bone marrow of the human body. A haematologist utilises microscopic investigation of the human blood, which encourages the need for techniques that are infinitesimal image shading, division, arrangement, and bunching, which permits the identification of a patient's experience with leukaemia. The manual infinitesimal assessment of bone marrow is less precise, tedious, and vulnerable to mistakes, which causes it to be hard for laboratory labourers to precisely perceive the qualities of impact cells. The additionally contrasted informational indexes and diverse shading models allow you to look at the presentation's changed shading pictures. Experts and professionals who work with stained photographs of leukaemia patients will benefit from this method's discovery of a crucial component. Algorithms for detecting leukaemia have been studied in this research. The methodology and efficiency are compared, which can be used for different analyses by the researchers.

COVID-19 has ushered in a new era of face masks dominating daily life. Furthermore, facial recognition is ineffective for medical personnel who wear surgical masks. Periocular biometrics is the automatic recognition and classification of a person based on features gathered from the area of the face that surrounds the eye. Analysed and identified various aspects from current work in this comprehensive survey work on periocular biometrics, such as datasets available for periocular regions, biometrics systems available, periocular area detection and segmentation, local and global descriptors, and so on. This research, as predicted, cites current and recent literature to demonstrate the shortcomings of periocular biometrics. This paper also outlined the direction of future periocular recognition studies.

Patient searches health information through online and other mode also. With different information they gather, it leads to enhance emotional intelligence and better self-management. This study aims to examine the literature on the dissemination of such information through web. It also explores the various approaches that scholars have used in this field. The article titles were extracted from the PubMed database using the keywords emotional intelligence, self-management, and health seeking behavior as the search terms. The study was conducted to identify the most influential journal articles published in the field of health seeking behavior. A total of 397 articles were retrieved using the keyword search. The research contributed to the study of bibliometric and offered a systematic assessment for health-related information seeking, self-management, and emotional intelligence publications published on the web. The aim of this study is to provide an understanding of the sources of information that can be used for research on this field of study.

Detailing is the method of communication in which a healthcare professional is conveyed about a product of a pharmaceutical industry. It is important to understand the preferences of physicians in the detailing process as physician have to cater to the treatment of patients and also keep themselves updated about the advancements made in the pharmaceutical industry. Hence, this study aims to highlight the preferences of physicians in the detailing process based on two different age groups of physicians, young and adult. Seven parameters: convenience of timing, day, interaction time, technology failure, duration of detailing, personalized information and social interaction skills of medical representatives were taken for the study. 425 physicians were approached, and their viewpoints were taken on a seven point Likert scale for the variables identified for the study. It was found that physicians above 45 years are more flexible in providing time to the medical representatives than physicians of age 25–45, and they expect more personalized information.

Promotion of prescription drugs to patients on social media is prohibited in India. However, online health websites have made this law a celebration. Patients are increasingly using the social network, particularly pharmaceutical social networking sites, to obtain health-related information. Viewpoints are disclosed in this article. This is a vital topic as differences in perception can lead to further disputes between patients and doctors. The survey includes 1500 patients and 400 doctors. The results show that there are significant differences in the perspective of doctors and patients. Because patients and doctors work together as a team, and patients are available to doctors for healthcare solutions, such large differences in their views on the pros of SMP can cause the partnership to fail.

Ghosal, Palash Kumar, Amish Kundu, Soumya Snigdha Srivastava, Utkarsh Prakash Datta, Ashis Sarma, vsThe coronavirus pandemic has caused one of the biggest global crises. With an inevitable need for fast screening of the disease, deep learning-based segmentation of Covid-19 infected lung regions in computed tomography (CT) scans gained significant attention. The automated screening procedure generated results significantly faster than the manual screening techniques and directly helped provide a wider outreach to patients. Therefore, to aid in computer-aided diagnoses, this paper presents AUTCD-Net (AUTomated framework for efficient Covid-19 Diagnosis-Network), based on hierarchical resolution steps, to efficiently segment Covid-19 infected lung regions in CT scans. The approach results in a 0.71 dice score and rivals all previous state-of-the-art approaches. The overall evaluation combined with our in-depth model analysis, and critical inferences can be further extended for developing a computer-aided diagnostic (CAD) tool to assist the CT image reading process for detecting Covid-19 infected regions in the near future.

Ghosal, Palash Kumar, Amish Datta, Ashis Sarma, Hiren KD Nandi, DebashisThe shifting of the brain midline, generally known as midline shift (MLS), is the dislocation of the brain from its actual position to a new position. Even though the shifting is very low, it has a worse effect on the overall body condition of the patient. The shifting measurement is critical for clinicians to understand the plan of action on an emergency basis. MLS of the brain can be measured by processing medical images like X-ray, ultrasound, computed tomography (CT), and magnetic resonance imaging (MRI). In this work, the main focus is given on CT images. CT scan of the head consists of large numbers of slices. All slices are not suitable for ideal midline detection. This paper gives a proposal for an automated computer-aided efficient midline estimation system with only two steps. The first step consists of selecting suitable slices and skull detection, while the second step is to separate soft tissues part inside the detected skull and draw the midline. A proposal is given to estimate the control points (centroid, most anterior and posterior points on the falx). Promising improvement concerning modified Hausdorff distance (MHD) and mean absolute distance (MAD) compared with some of the existing methods was also observed.

In this paper, we have implemented a deep neural network for classification of ASD and neuro typical subjects, based on functional connectivity features derived from fMRI data. We have derived functional connectivity features from resting-state fMRI data downloaded from ABIDE. Further, we investigated the variation of the model performance for different brain atlases.

Automated Patient monitoring is rising to importance in the mobile healthcare services as it makes day-to-day activities risk-free, by continuously monitoring their vital signs. Clinical solutions are being provided to patients in no time, which is made possible due to the latest improvements in the “Internet of Things (IoT), cloud computing, and fog computing”. “Machine learning and Deep learning” are now being extensively used for various applications in healthcare such as extracting relations from vast amounts of patient data, analyzing patterns to predict the propagation of diseases, classify reports and X-rays to detect diseases, to name a few. In this paper, a deep learning-based model is proposed to monitor Covid-affected patients within hospitals. Our model can provide an online link between a patient and medical facility while also collecting patient data. This will enhance the care taken for patients. At the hospital end, we present a deep learning model using ResNet-50 that could classify chest X-rays as Covid positive or No Covid. Through this model we expect to quicken the process of COVID-19 detection while lowering the healthcare expenses.

The main function of a recommendation engine is to filter the data and to use algorithms and to suggest the items which is more applicable to the users. It also identifies the past behaviour of the customer and planted on products that the users want to buy most. Now when a new user visits their e-commerce site, then it will not show any past of that new user. Now question is how that site come to know about the recommended products to the new user. There will be a solution that the bestselling product of that e-commerce site is the highly demanded products. There will be another solution that could be the product which gave the maximum product to the business to that e-commerce site. In this present work, we have created a model for popularity-based recommendation system using Machine Learning. Finally, we got top 5 popular products which are recommended for users using KNN algorithm.

Netflix is a subscription-oriented streaming service platform which is available on various application stores that allows customers or viewers to view television series and movies through any Internet-connected device (irrespective of small screen or large screen) without any advertisement interruptions. Since Netflix has such a large collection of content, it can be difficult to decide which movie or show to watch. Users on the Internet are always looking for something new and something that they like, and to do so, they search endlessly on the Internet. If the time spent searching does not yield a fruitful result, all the time spent searching will be wasted, and the entire purpose of being on an “over the top” platform that promotes instant playback so that people can watch anything at any time will be futile. And, if the user is new to the platform, he or she will take some time to become acquainted with the environment, such as the variety of content, his or her area of interest, and following up with a specific content. A recommendation system can solve the problem of searching by intelligently filtering a large volume of dynamically generated data to provide customised services. It can assist people in quickly finding the movie or show of their choice. In this work, a movie and show recommendation system have been developed for Netflix services using the support vector machine learning approach. The emphasis has been given on developing a simple recommendation system which provides a good user experience. It is found that the system can recommend a list of movies or shows based on the user’s searching interest, thereby saves time.

SDN is a three-layered architecture with a centralized controller. Due to the increased usage of the Internet, there is a rapid generation of data traffic. Identifying and classifying traffic for effective network management is of prime importance today. SDN with deep learning (DL) model is a key area that has achieved high recognition. Moreover, in order to identify and classify network data traffic, the classification model needs to achieve high accuracy. This will ease network management and resource allocation. There exist many DL models for traffic classification but in this research, long short-term memory (LSTM) is studied, and insights are highlighted. Survey concludes that LSTM model for traffic classification is being used in the major area of data security and intrusion detection. Delay in classifying network traffic needs a special game theory approach which could not be handled by many LSTM models.

This article presents an innovative visual aid framework for completely blind people, which takes the form of a pair of glasses. The following are some of the most essential characteristics of the proposed device. The complicated algorithm processing is carried out on the Raspberry Pi 3 Model B+, which has low-end computing power. Using a combination of camera and ultrasound sensors and GPS-based location tracking for use in a navigation system, this Internet of things-based device offers advanced dual detection and distance measurement capabilities. This device makes it possible to have better access, solace, and navigational ease to blind people.

Twitter is an excellent initial point for social media analysis. People directly share their opinions through Twitter with the public. One of the very common analyses which can perform on many tweets is sentiment analysis. The 2020–2021 Indian farmer’s protest is a protest against three farmer acts which were passed in parliament of India in 2020. In this paper, we have performed sentiment analysis of the protest of farmers in India (2020–2021) by considering the opinion of the people. The data is taken from hashtags that are related to farmers protest and some minor hashtags related to farmers. Based on the result of analysis, we conclude the impact of the protest done to repeal the farmer’s act, on India. On analysis, we obtained the result as percentages of positive, negative and neutral.

In this present age, everyone uses social media. Whilst ideas spread through these mediums can be colourful, there are certain views of people which are shared with a tone of negativity, mischief and offensiveness to them as well. One of the most used of such social media platforms include Twitter, which is a platform where messages with limited word limit are sent by the people. This present work, thus, is about analysing the polarity of a tweet by using millions of data achieved through Natural Language Processing (NLP). We collected a large dataset of tweets which are marked by levels of polarity from negative to positive. We use Machine Learning algorithms on this dataset to detect the polarity of the tweet. Here we have used Multinomial Naïve-Bayes, Complement Naïve-Bayes and Logistic Regression classifier to find the polarity of tweets. The dataset size is 1.6 million and we got the best result using Logistic Regression. The highest accuracy is 78.05%.

Social media promotion of prescription medicines are subject to ethical and legal standards. Contacting patients directly to promote drugs has long been frowned upon and banned in all but two countries: the United States and New Zealand. Technology has made this sensible approach obsolete. Patient’s interest in searching social media for health information is increasing rapidly, including information from drug marketers. This development was praised by many groups. The reason for this is that it offers many rewards. Social media promotion (SMP) offers well-known advantages. The purpose of this study is to assess the association of the positive aspects of SMP (such as emotional support and medical adherence) with SMP. Our results suggest that patients rate SMP very much associated with emotional support and medical adherence, which has important implications for marketers, clinicians, and policy makers who are critical in promoting the development of safe systems to improve health systems. Additionally, future studies should use an advanced communication model adapted to the social media environment to ensure thorough research.

Video description is basically a method of automatic creation of natural language sentences which can illustrate the insides of a given video. Several applications areas are there where we need summarization of description of video such as in video subtitling, human–robot interaction, helping the visually impaired person etc. In this present paper, we review different state-of-the-art techniques for video description using deep learning techniques. This paper summarizes the deep learning based state-of-the-art methods for video description. The paper also identifies the pros and cons of each evaluation metrics employed in the task. Most of these methods generally uses a 3D CNN model to convert videos into multi-dimensional arrays, then uses a word embedding techniques like GLOVE, etc., for featurization of text description and then finally trains an RNN or LSTM model or a variant of the two to perform final classification. The paper also describes the benchmark datasets of each of these methods along with evaluation metrics and state-of-the-art performance is reported on the same. Wherever applicable, we also list down the advantages and drawbacks of each of these methods as stated earlier. This survey paper can act as a preparatory read for anyone entering the field.

Image classification task has been an important area in the field of computer vision study as it is applied in varied applications. The approaches for image classification is based on feature selection of image class and effectively applied through various low level feature algorithms for matching with a particular class and yielding a classification result identifying one from another. The task for feature selection particularly become extremely challenging to spatial images such as aerial images, as they contain varied complex feature, scale challenges as well as image orientation. A more accepted approach of image classification for aerial image is the use of Convolution Neural Networks (CNN). CNN models are capable of producing higher accuracy compared to contemporary feature based algorithm as they tend to utilize higher local features. Transfer learning approach using readily available and proven CNN for image classification makes the task a step closer to capturing and designing a CNN which adapt to user dataset and classification requirements. There is however the need to create user dataset with sufficient images to retrain the network for fine-tuning and testing its accuracy. This paper presents such experiment using GoogLeNet pre-trained network which is subject to replacing of the fully-Connected layer and output layer as per the classification requirement. The network is further subject to training and testing using test dataset other than that have never been exposed to the network. An accuracy of 98.33% was achieved.

Borah, Naiwrita Baruah, UdayanSince the emergence of the Internet and low-cost digital image sensors, the number of image databases has grown tremendously. These image databases must have efficient image retrieval methods. One such technique is content-based image retrieval. In these databases, one will find charts, graphs, pictures, and even some text. Our main focus is on visual information retrieval by using this data. The idea is to bridge the semantic gap of high-level human perceptions and low-level features. This review was conducted based on an assessment and comparison of the most current CBIR approaches. Machine learning algorithms, similarity matching techniques, and performance assessment methodologies are also included in this study. This study provides an in-depth look into CBIR, covering its theory, concepts, techniques, difficulties, future directions, and performance.

Education sector is fastest growing as well as major contributing sector in society. Although like every sector in society, education is also facing issues in term of security. Since the pandemic rise and resulting lockdowns, education industry is restricted learning to remote access on virtual platforms and providing the combination of physical classroom as well as virtual training. Whilst this is of enormous help to students and educational institutes, it does come with cyber risks. Education sector has not been exempted from the inevitability of cyber risks and threats that prevail in the dark side of the booming technological developments. Hence, it has become imperative to not only identify these risks but also classify them in order to rectify and understand their consequences. Therefore, in this paper, we will describe the vulnerable areas in educational organisation and predict the attacks according to the impact of the attack which is part of the risk mitigation, so that they can be detected and prevented by using machine learning (ML).

Air pollution, in general, relates to the introduction of toxins into the atmosphere that is toxic to people’s health and therefore the entire ecosystem. It has the potential to be one of the foremost dangerous risks mankind has ever encountered. It hurts cattle, livestock, and trees, among other things. To avoid this issue, machine learning (ML) algorithms must be used to forecast air quality (AQ) from pollution in the transportation field. Therefore, AQ measurement and forecasting have become a significant research subject. Here in this work, we aimed to look at ML-based approaches for AQ forecasting with the highest degree of accuracy. The entire dataset will be evaluated using the supervised ML technique to collect multiple pieces of information such as variable recognition, univariate analysis, bivariate and multivariate analysis, missed value treatments and information confirmation, information cleaning/preparing, and visualization. Our study offers a detailed guide to model parameter sensitivity analysis in terms of results in AQ emissions prediction through accuracy measurement. To suggest an ML-based approach for reliable forecasting of the air quality index (AQI) value by comparing supervised classifier findings in the form of better accuracy. To suggest a ML-based approach for reliably forecasting the AQI value by evaluating supervise classification, ML algorithm prediction outcomes in the form of best accuracy. By predicting AQI, we can recall the major factors that cause pollution and the area affected by pollutants in various places in India.

Nature never fail to inspire us; we can learn a lot from them. Based on the inspiration acquired from the biological activities of nature, bio-inspired algorithms are developed to compete with modern rival techniques. An optimal solution can be obtained for intricate engineering and scientific problems when a bio-inspired algorithm is embedded with machine learning techniques. The problems are proposed with multiple nonlinear constraints which require huge time and high dimensionality. In this overview, various bio-inspired algorithms like artificial bee colony (ABC) algorithm, fish swarm algorithm (FSA), cat swarm optimization (CSO), whale optimization algorithm (WOA), artificial algae algorithm (AAA), elephant search algorithm (ESA), chicken swarm optimization algorithm (CSOA), moth flame optimization (MFO), and gray wolf optimization (GWO) algorithm are discussed precisely.

In the software sector, there is a wide range of painting software. Each of these painting applications has its own set of benefits and drawbacks. The common disadvantage in all these programs is that each of them requires a pointing device to draw. Our project is centered on the creation of a system that draws on a screen using human hand gestures. The user doesn’t have to make use of any external hardware device to draw. This system makes use of different hand gestures to draw on the screen. This survey paper presents a comparison of various methodologies proposed by different authors on “hand gesture-based drawing” system. These systems make use of Human Hand Gestures for writing characters, words, or free drawing in free space. There are systems that make use of a stylus, digital pen, electronic hand gloves, and other external equipment to interact with computers and there are certain systems that do not make use of such external equipment but, make use of only hand gestures to interact with computers. Each of these methods has its own advantages and disadvantages. The aim of this survey paper is to present the most efficient and accurate method out of the below-elaborated methodologies.

Changing daily load, renewable integration, transmission losses and one-way communication are the major challenges of conventional grid. These problems are solved by introducing smart grid by integrating digital and fast data computing system with convention grid. In smart grid, bi-direction communication and intelligent data analysis plays important role in enhancing operational cost and efficient power supply. Short-term electric load forecasting using past data by the modern computational techniques provides great helps in minimize the losses and demand side management. The paper brief about modeling and simulation of electrical load using different machine learning techniques like random forest model, gradient boosted trees, decision tree, and generalized line. Finally, the model having minimum error can use for accurate forecasting of future electrical load. This helps to minimizing the losses and enhance stability of the grid.

Paul, Shubhrojit Handique, Mousum Sarma, Hiren Kumar DevaThe basic feature of a computational machine which makes it easy to perform the tasks depends largely on the computational power. The recent advancement of quantum computation has gained importance because of its potential use of increasing computational power and high efficiency of information processing. In general, finding two prime numbers of a very large number in a classical computation is a laborious task, but the quantum computation shows the evidence of fast computation to solve this problem. Due to the intrinsic properties of quantum computation, it is different from classical computers. In this paper, we specifically deal with quantum circuits and its faults. Here, quantum circuits involve the computational routine of quantum operations on quantum data such as qubits and real-time classical computation. In similar to the conventional circuit, the testing procedure is needed for observing the correct behaviour on qubits in quantum circuits. In this paper, we basically deals with single missing gate fault (SMGF) that might occur in quantum circuits. For this purpose, we proposed a fault detection method for SMGFs and experimental analysis is performed.

Satellite signal data for the complete year of 2019 with the antennas of frequencies 20.2 GHz and 30.5 GHz in the region of Shillong, India has been examined for frequency scaling of attenuation found in existing literature. Laser precipitation monitor (LPM) data was recorded at the rate of one sample per minute for rain rate and signal data was recorded at the rate of one sample per second for the frequencies of 20.2 GHz and 30.5 GHz. The deviation of experimental result from the existing model has been analyzed and error performance has been observed.

With the advancement in technology and the fast-growing AI, intelligent systems are introduced every day. Detection and acknowledgment of an item from a real-time image independent of their scale, and the view is troublesome. It turns out to be significantly more basic and fundamental when managing complex regions like schools, offices, factories, etc. In this work, we introduced and developed An school uniform dress identification using deep learning based (SUIUDL) Approach. The SUIUDL performs object classification based on the uniform dress convolution neural network (CNN), and high-performance computing system in a real environment. The SUIUDL targets and identifies the special features of the school uniform and apply the deep learning technique of convolution neural network (CNN) for classifications of the dress.