Browse Items (2150 total)
Sort by:
-
Analysis of Challenges Experienced by Students with Online Classes During the COVID-19 Pandemic
In the current context of the COVID-19 pandemic, due to restrictions in mobility and the closure of schools, people had to shift to work from home. India has the worlds second-largest pool of internet users, yet half its population lacks internet access or knowledge to use digital services. The shift to online mediums for education has exposed the stark digital divide in the education system. The digitization of education proved to be a significant challenge for students who lacked the devices, internet facility, and infrastructure to support the online mode of education or lacked the training to use these devices. These challenges raise concerns about the effectiveness of the future of education, as teachers and students find it challenging to communicate, connect, and assess meaningful learning. This study was conducted at one of the universities in India using a purposive sampling method to understand the challenges faced by the students during the online study and their satisfaction level. This paper aims to draw insight from the survey into the concerns raised by students from different backgrounds while learning from their homes and the decline in the effectiveness of education. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
Social Media Sentiment Analysis of Stock Market on Tweets
Sentiment categorization is utilized in today's world to analyze social media data about the stock market and estimate its future stock movement. We investigated the possible influence of 'public sentiment' on 'market trends' using sentiment analysis and machine learning concepts. Due to the enormous number of components involved, such as economic situations, political events, and other environmental factors that may affect the stock price, stock price prediction is an exceedingly complicated and challenging process. Because of these considerations, evaluating a single factor's influence on future pricing and trends is challenging. As more individuals spend time online, the popularity and impact of numerous social media platforms has skyrocketed in recent years. Twitter is one such social media tool that has exploded in popularity. Twitter is a terrific place to stay up to speed on current societal trends and perspectives. The 'Twittersphere' is a melting pot that supports diverse viewpoints, emotions, and trends, and it has the potential to be a crucial influencer in influencing and shaping perceptions. 2022 IEEE. -
Analysis of Fine Needle Aspiration Images by Using Hybrid Feature Selection and Various Machine Learning Classifiers
Women die of breast cancer most often worldwide. Breast tissue samples can be examined by radiologists, surgeons, and pathologists for evidence of this cancer. Fine needle aspiration cytology (FNAC) can be used to detect this cancer through a visual microscopic examination of breast tissue samples. This sample must be examined by a cytopathologist in order to determine the patient's risk of breast cancer. To determine if a tumor is malignant, the nuclei of the cells must be characterized by their chromatin texture patterns. A machine learning method is used in order to categorize FNA images into two classes, respectively Malignant and Benign. For detecting abnormalities, numerous feature collection methods and machine learning means are applied here. Using features extracted from the FNA image set, UCI machine learning datasets are used to validate the proposed approach. This paper compares three classification methodologies, namely random forests, Naive Bayes, and artificial neural networks, by examining their accuracy, specificity, precision, and sensitivity, respectively. With the ANN and PCA along with the Chi-square selection method, 99.1% of the classifiers are correctly classified. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
An Improved and Efficient YOLOv4 Method for Object Detection in Video Streaming
As object detection has gained popularity in recent years, there are many object detection algorithms available in today's world. Yet the algorithm with better accuracy and better speed is considered vital for critical applications. Therefore, in this article, the use of the YOLOV4 object detection algorithm is combined with improved and efficient inference methods. The YOLOV4 state-of-the-art algorithm is 12% faster compared to its previous version, YOLOV3, and twice as faster compared to the EfficientDet algorithm in the Tesla V100 GPU. However, the algorithm has lacked performance on an average machine and on single-board machines like Jetson Nano and Jetson TX2. In this research, we examine the performance of inferencing in several frameworks and propose a framework that effectively uses hardware to optimize the network while consuming less than 30% of the hardware of other frameworks. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
A Comparative Investigation on the use of Machine Learning Techniques for Currency Authentication
In the present banking sector, identifying the real and the fake note is a very challenging task because if we do it manually, it takes a long time to check which is real and which is fake. This research study article aims to authenticate the money between real and fake by using different machine algorithms facilitating learning, such as K-means Clustering, Random Forest Classification, Support Vector Machines, and logistics Regression. Specifically, we consider the banknote dataset. The data of money is extracted from various banknote images by using the wavelet transform tool, which is primarily used to remove elements from the images. However, we are mainly concerned with the different machine learning algorithms, so we take the two variables, where the first variable indicates image variance and the second indicates image skewness. We use these two variables to train our machine learning algorithms. So, majorly, by applying the different machine learning algorithms, which are supervised and unsupervised, we find the accuracy for the respective machine learning algorithms and then visualize and classify the real and fake notes separately. Finally, the prediction is based on integrity, which means the efficiency value is based on how much the mechanism system can uncover the fake notes. Then, after calculating the accuracy of currency authentication, there is a high possibility that the accuracy of the particular algorithm is the best algorithm, so the application of currency authentication will be very useful for the bank to easily find duplicate notes. 2022 IEEE. -
Hybrid AI Talent Acquisition Model: An Opinion Mining and Topic based approach
Artificial Intelligence models have found their usage in the human resource domain. In this paper, job reviewers' opinions on online discussion boards have been captured. The relative importance of factors has been established through an extensive literature review. First, LDA Topic modelling by adopting PCA is performed on unstructured text data has been analyzed. Second, sentiment analysis using the Li-Hu method has been employed to understand job seekers' satisfaction with job portals. The proposed model, 'Hybrid AI Talent Acquisition Model,' follows a novel approach to streamlining the jobseeker opinion related to online outlets. 2022 IEEE. -
Pulse Shaper Design for UWB-Based Medical Imaging Applications
In this paper, a pulse shaping filter is designed to shape the higher-order derivatives of the basic UWB Gaussian pulse for efficient pulse transmission through human tissues for medical imaging applications. The shaped pulse for the desired center frequency fits the FCC mask and power spectral density (PSD) specifications with higher spectral efficiency being achieved. It is observed that the ringing effect of Gaussian pulse is reduced by using the proposed bandpass FIR shaping filter. The low ringing effect observed in the shaped pulse ensures better antenna power distribution and improved location accuracy which is critical factor for medical imaging applications. The pulses synthesized are highly orthogonal which aids in multi-access communication, improved bit error rate (BER) performance and short duration UWB pulses leading to higher data rate transmission. The drooping frequency response characteristics of the synthesized pulse have reduced clutter hence tightly focused image obtained for imaging applications. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
Machine Learning Methods leveraging ADFA-LD Dataset for Anomaly Detection in Linux Host Systems
Advancement in network technology and revolution in the global internet transformed the overall Information Technology (IT) infrastructure and its usage. In the era of the Internet of Things (IoT) and the Internet of Everything (IoE), most everyday gadgets and electronic devices are IT-enabled and can be connected over the internet. With the advancements in IT technologies, operating systems also evolved to leverage these advancements. Today's operating systems are more user-friendly and feature-rich to support current IT requirements and provide sophisticated functionalities. On the one hand, these features enabled operating systems accomplish all current requirements, but on the other hand, these modern operating systems increased their attack surface considerably. Intrusion detection systems play a significant role in providing security against the broad spectrum of attacks on host systems. Intrusion detection systems based on anomaly detection have become a prominent research area among diverse areas of cyber security. The traditional approaches for anomaly detection are inadequate to discover the operating system level anomalies. The advancement and research in Machine Learning (ML) based anomaly detection open new opportunities to tackle this challenge. The dataset plays a significant role in ML-based system efficacy. The Australian Defence Force Academy Linux Dataset (ADFA-LD) comprises thousands of normal and attack processes system call traces for the Linux platform. It is the benchmark dataset used for dynamic approach-based anomaly detection. This paper provided a comprehensive and structured study of various research works based on the ADFA-LD for host-based anomaly detection and presented a comparative analysis. 2022 IEEE. -
Customer Behavior Analysis Using Unsupervised Clustering and Profiling: A Machine Learning Approach
Now-a-days, client conduct models are reliably established on information mining of client information, and each model is supposed to answer one solicitation at one point on schedule. Anticipating client conduct is a problematic and irksome task. Thus, making client conduct models requires the right strategy and approach. Right when an estimate model has been fabricated, it is challenging to restrict it for the motivations driving the advertiser, to pick the very thing displaying moves to make for every client or for the party of clients. Notwithstanding the multifaceted nature of this arrangement, most client models are completely fundamental. As the need might arise, most client conduct investigation models ignore such endless proper factors that the gauges they make are overall not altogether strong. This paper plans to encourage a connection rule mining model to expect client conduct using a typical electronic retail store for data combination and concentrate critical examples from the client conduct data. In this undertaking, a solo grouping of information on the customer's records from a regular food item company's data set will be played out. Customer segmentation is the act of clustering customers into bunches that reflect likenesses among customers in each group. Customers are separated into sections to advance the meaning of every customer to the business. To change items as indicated by unmistakable requirements and practices of the customers. It additionally assists the business with obliging the worries of various kinds of customers. Customers were clustered using a technique known as agglomerative clustering, which is a type of hierarchical clustering. Agglomerative clustering is a method for clustering data in a hierarchical order. It entails merging cases until you reach the appropriate number of clusters. The number of clusters to be produced is determined using the Elbow Method. 2022 IEEE. -
A Survey on Adaptive Authentication Using Machine Learning Techniques
Adaptive authentication is a reliable technique to dynamically select the best mechanisms among multiple modalities to authenticate a user based on the users risk profile generated using behavior and context-based information. Websites or enterprise applications enabled with adaptive authentication will have a more robust security system as analyzing the large volume of the user, device, and browser data in real time generates a risk score that decides the appropriate level of security. Though a significant amount of research is being carried out on adaptive authentication, no single model is suitable for a global attack. This paper provides a structured (extensive) survey of current adaptive authentication techniques available in the literature to identify the challenges which demand future research. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
Improvement to Recommendation system using Hybrid techniques
Currently, recommendation systems are a common tool for providing individualized recommendations and item information to users. For personalization in the recommendation system, there are a variety of strategies that can be used. To improve system performance and offset the shortcomings of individual recommendation strategies, a hybrid recommender system integrates two or even more recommendation techniques. The demand to summarize all of the knowledge on actual methods and algorithms utilized in hybrid recommended systems necessitates the need for a systematic review in the domain. These materials will be employed to aid in the development of an auto-switching hybrid recommender system. In the content-based filtering technique, the algorithm is based on the contents of items and the collaborative filtering technique algorithm combines the relationship between user and item. Both of the approaches of recommendation system are suffers from some limitations, this is a big issue to predict better recommendations to the user. Hybrid systems are introduced to overcome the main limitations of both techniques. These systems are made with a combination of content-based and collaborative filtering techniques and have advantages of both techniques. With the use of hybrid systems, the quality of recommendations is improved. Hybrid recommendation systems use previous data of a user to find his/her interest and then they target the set of an adjacent user which is similar with that user and according to adjacent user recommend things to the user. Hybrid systems offer the items that share the common things that a user rated highly (Content-based filtering) and make suggestions by comparing the interest of a similar user (Collaborative filtering). 2022 IEEE. -
Machine Learning Techniques in Predicting Heart Disease a Survey
The heart serves an important role in living creatures. Diagnosis and forecast of cardiac illnesses demand greater precision, perfection, and accuracy because such tiny mistakes can lead to weariness and death. Numerous heart-related deaths have occurred, and the incidence rates have been rising over time. Predicting the development of heart disorders is important to work in the medical industry. Every month, many databases related to the patient are kept. The information gathered can be used to predict the occurrence of future diseases. This article gives an outline of cardiovascular diseases and modern treatments. Also, the focus of this research is to outline some current research on applying machine learning techniques to predict heart disease, analyze the many machine learning algorithms employed, and determine which technique(s) are useful and efficient. Artificial neural network (ANN), decision tree (DT), fuzzy logic, K-nearest neighbor (KNN), Naive bayes (NB), and support vector machine (SVM) are data mining and machine learning approaches used to predict cardiac disease. This paper includes an overview of the present method based on features, the algorithms are compared, and the most accurate algorithm is analyzed. 2022 IEEE. -
CNN-RNN based Hybrid Machine Learning Model to Predict the Currency Exchange Rate: USD to INR
Foreign currency exchange plays an imperative part in the global business and in monetary market. It is also an opportunity for many traders as an investment option and the advance knowledge of fluctuation helps the investors making right decision on time. However, due to its volatile nature, prediction of foreign currency exchange is a challenging task. This paper implements two models based on machine learning, namely Recurrent Neural Networks (RNN) and a Hybrid model of Convolutional Neural Networks (CNN) with RNN known as CNN-RNN to assess the accuracy in predicting the conversion rate of US Dollar (USD) to Indian Rupees (INR). The data set used to verify and validate the models is the daily currency exchange rate (USD to INR) available in public domain. The experimental results show that the simple RNN model performs slightly better than the hybrid model in this particular case. Though the accuracy of the hybrid model is very high in terms of error calculation still the single RNN model is the better performer. This does not straight away reject the hybrid model rather needs more experimental analysis with changing architecture and data set. 2022 IEEE. -
An Analytical Study on the influence of using Trimmed Gait Energy Images for Human Gait Biometrics using Deep Learning
Gait based human recognition is founded on the principle that every human being has a distinctive style of walking. With the rise in the use of video surveillance devices, gait is one of the most convenient biometrics to use, in forensics. This paper is an analytical study of the effect of using trimmed Gait Energy Images (GEI) for Human Recognition using different deep learning techniques. Gait energy images are a spatiotemporal, silhouette-based representation of the human gait. GEIs from the CASIA B Multiview dataset was used to build two other sets of data by subtracting the upper body Deep learning and transfer learning techniques including Convolution Neural Networks (CNN) and VGG16 algorithms had been implemented to carry out the recognition. Results showed that the performance of the model using upper body images gives a greater accuracy than the lower body images. It has also been observed that the accuracy of recognition provided by the upper part of the body is almost the same as that achieved by the whole body, which brings forth the idea that the upper part of the body is the most pertinent in Human Identification using Gait as a biometric. 2022 IEEE. -
Machine Learning based Loan Eligibility Prediction using Random Forest Model
When one or more people, organizations, or other entities lend money to other people, organizations, or entities, it is known as a loan. The recipient (that is the borrower) takes on a debt for which he or she is normally accountable for interest payments until the loan is repaid. The major goal of this proposed model is to ensure that an individual, institution, or organization seeking for a loan is properly verified before granting them the loan they require. Before authorizing a loan for any individual or business several factors must be considered. That including gender, education, and the number of dependents. The goal of proposed model is to automate the method, which will save time and energy while improving the efficiency of the process. This particular process input is having two different kind of data set. First one is train data set and second set is test data set. The first date set that is train data set is generally used to train and assess the machine learning model accuracy. The loan eligibility predictions are generated using the test data set. To forecast loan eligibility and train this random forest, machine learning method called Random Forest. The proposed random forest model is providing higher accuracy level. This model is providing 28 % higher accuracy level compare to regular prediction. 2022 IEEE. -
CoInMPro: Confidential Inference and Model Protection Using Secure Multi-Party Computation
In the twenty-first century, machine learning has revolutionized insight generation by using historical data across domains like health care, finance, and pharma. The effectiveness of machine learning solutions depends largely on the collaboration between data owners, model owners, and ML clients, without privacy concerns. The existing privacy-preserving solutions lack efficient and confidential ML inference. This paper addresses this inefficiency by presenting the Confidential Inference and Model Protection, also known as the CoInMPro, to solve the privacy issue faced by model owners and ML clients. The CoInMPro technique is suggested with an aim to boost the privacy of model parameters and client input during ML inference, without affecting the accuracy and by paying a marginal performance cost. Secure multi-party computation (SMPC) techniques were used to calculate inference results confidentially after sharing client input and model parameters privately from different model owners. The technique was implemented in Python language using the open-source SyMPC library to support the SMPC function. The Boston Housing Dataset was used, and the experiments were run on Azure data science VM using Ubuntu OS. The result suggests CoInMPros effectiveness in addressing privacy concerns of model owners and inference clients, with no sizable impact on accuracy and trade-off. A linear impact on performance was noted with an increase of secure nodes in the SMPC cluster. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
One Time Password-Based Two Channel Authentication MechanismUsing Blockchain
Using Fog Nodes, also known as IOT devices are increasing everyday with more and more home automation, industry automation, automobile automation, etc. Security threats for these devices are also increasing. One of the threats is impersonating one fog node, stealing data and taking control of the network which is also known as the Sybil attack. To provide security, most fog devices use one step or two step authentication and sometimes use encryption. With static passwords, there is a chance of compromise by password sharing and leaking. Some weak encryption algorithms used are also compromised. Data about fog nodes in the network is stored in a weak database and is tampered. OTP-based Two Channel Authentication Mechanism (OTPTAM) to authenticate the fog nodes with metadata stored in Blockchain Database and communicate using channels encrypted with Elliptical Ciphers can solve the majority of these problems. Metadata of the nodes like Bluetooth MAC address, network mac address, telephone number are all stored in the blockchain and the OTP is exchanged via these channels to ensure the authenticity of the fog nodes. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
An Efficient Face Recognition System using Deep Transfer Learning
Face recognition is an AI-based innovation used to find and recognize human appearances in videos and images. Organizations can apply face recognition to many different kinds of fields which may include biometrics, regulation of law, security and individual wellbeing; so as to take observation of individuals in any scenario. Face recognition has advanced from simple vision methods to progress in ML; and further to progressively refined neural networks (ANN) and related advances. It currently assumes an indispensable part as the initial phase in numerous basic applications, including the task of tracking a face. Face recognition is utilized to focus cameras or count the number of individuals present in a particular region. The innovation likewise has showcasing applications, for instance, showing recommended promotions when a specific user is detected. 2022 IEEE. -
A Multi Objective Artificial Eco-System Based Optimization Technique Integrating Solar Photovoltaic System In Distribution Network
Agricultural sector contributes 6.4% of total economic generation across the world. Notably, the utilization of technology to improve the yield and economy is rapidly increasing. To provide continuous supply to the residential customers, the agricultural feeder grid-dependency has to be integrated with Solar Photo Voltaic (SPV) systems. In this paper, an Artificial Eco-System based Optimization (AEO) algorithm is proposed for simultaneously identifying the locations and quantifying the sizes of SPV systems. A practical distribution system feeder 'Racheruvu 11kV agricultural feeder' Andhra Pradesh, India is considered for simulation purpose and the performance is compared with the standard IEEE-33 radial distribution system. 2022 IEEE. -
Applying Ensemble Techniques for the Prediction of Alcoholic Liver Cirrhosis
More than fifty percent of all liver cognate deaths are caused by alcoholic liver disease (ALD). Excessive drinking over the time leads to alcohol-related steatohepatitis and fatty liver, this in turn can lead to alcoholic liver fibrosis (ALF) and in due course alcohol-related liver cirrhosis (ALC). Detecting ALD at an early stage will reduce the treatment cost to the patient and reduce mortality. In this research, a two-step model is developed for predicting the liver cirrhosis using different ensemble classifiers. Among 41 features recorded during data collection, only 15 features arefound to be effective determinants of the class variable. The proposed stacked ensemble technique for ALD prediction is compared with other ensemble models such as random forest, AdaBoost, and bagging. Through experimentation, it is observed that the proposed model with XGBoost and decision tree as base models and logistic regression as Meta model exhibits prediction accuracy of 93.86%. The prediction accuracy of theproposed stacked ensemble technique is 0.2% better in prediction accuracy and 0.3% reduced error rate in comparison with random forest classifier. 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.