Browse Items (11855 total)
Sort by:
-
A Comprehensive Research on Deep Learning Based Routing Optimization Algorithms in Software Defined Networks
Discovering an optimal routing in Software Defined Networks (SDNs) is challenging due to several factors like scalability issues, interoperability, reliability, poor configuration of controllers and security measures. The compromised SDN controller attacks at the control plane layer, packet losses in the topology and end-to-end delay are the most security risk factors in SDNs. To overcome this, in most of the existing researches, Deep Reinforcement Learning (DRL) algorithm with various optimization techniques was implemented for optimal routing in SDN by providing link weights to balance the end-to-end delay and packet losses. DRL used Deterministic Policy Gradient (DPG) method which acts as an actor-critic reinforcement learning agent that searches for an optimal policy to minimize the expected cumulative long-term reward. However, discovering an optimal routing with efficient security measures is still a major challenge in SDNs. This research proposes a detailed review of routing optimization algorithms in SDN using Deep Learning (DL) methods which supports the researchers in accomplishing a better solution for future research. 2023 IEEE. -
Performance Analysis of Deep Learning Pretrained Image Classifiation Models
Convolutional Neural Networks (CNNs) is revolutionized in the field of computer vision, with the high accuracy and capability to learn features from raw data. In this research work focused on a comparative analysis of two popular CNN architectures, VGG16 and VGG19. The CIFAR dataset consists of 60,000 images, each with a resolution of 32x32 and it's belong to one of the 10 classes. Experimental results are compared with VGG16 and VGG19 in terms of their accuracy and training time, and to identify any differences in their ability to learn features from the CIFAR-10 dataset. The results of this research can aid in directing the choice of appropriate architectures for image classification tasks as well as the advantages of optimisation strategies for enhancing the efficiency of deep learning models. In order to enhance the performance of these structures, more optimisation methods and datasets may be investigated in subsequent research. 2023 IEEE. -
Artificial Intelligence Based Enhanced Virtual Mouse Hand Gesture Tracking Using Yolo Algorithm
Virtual mouse technology has revolutionized human computer interaction, allowing users to interact with digital environments without physical peripherals. The concept traces back to the late 1970s, and over the years, it has evolved with significant advancements in computer vision, motion tracking, and gesture recognition technologies. In recent times, machine learning techniques, particularly YOLOv8, have been integrated into virtual mouse technology, enabling accurate and swift detection of virtual objects and surfaces. This advancement enhances seamless interaction, intuitive hand gestures, and personalized virtual reality experiences tailored to individual user preferences. The proposed model, EHT (Enhanced Hand Tracking), leverages the power of YOLOv8 to address the limitations of existing models, such as Mediapipe. EHT achieves higher accuracy in hand tracking, real-Time hand gesture recognition, and improved multi-user interactions. It adapts to users' unique gestures over time, delivering a more natural and immersive computing experience with accuracy rates exceeding those of Mediapipe. For instance, across multiple sample datasets, EHT consistently outperformed Mediapipe in hand tracking accuracy. In Sample Dataset 1, EHT demonstrated an accuracy of 98.3% compared to Mediapipe's 95.65%. Similarly, in Sample Dataset 2, EHT achieved an accuracy of 99.35%, surpassing Mediapipe's 94.63%. Even in Sample Dataset 3, EHT maintained its superiority with an accuracy of 98.54 %, whereas Mediapipe achieved 98.26%. The successful implementation of EHT requires a custom dataset and optimization techniques to ensure efficiency on virtual reality hardware. EHT model is anticipated redefining how users interact with digital environments, unlocking new possibilities for intuitive and immersive computing experiences. 2023 IEEE. -
Malicious Traffic Classification in WSN using Deep Learning Approaches
Classifying malicious traffic in Wireless Sensor Networks (WSNs) is crucial for maintaining the network's security and dependability. Traditional security techniques are challenging to deploy in WSNs because they comprise tiny, resourceconstrained components with limited processing and energy capabilities. On the other hand, machine learning-based techniques, such as Deep Learning (DL) models like LSTMs, may be used to detect and categorize fraudulent traffic accurately. The classification of malicious traffic in WSNs is crucial because of security. To protect the network's integrity, data, and performance and ensure the system functions properly and securely for its intended use, hostile traffic categorization in WSNs is essential. Classifying malicious communication in a WSN using a Long Short-Term Memory (LSTM) is efficient. WSNs are susceptible to several security risks, such as malicious nodes or traffic that can impair network performance or endanger data integrity. In sequential data processing, LSTM is a Recurrent Neural Network (RNN) appropriate for identifying patterns in network traffic data. 2023 IEEE. -
Deep Learning Based Face Recognized Attendance Management System using Convolutional Neural Network
In today's digital age, manual attendance tracking is plagued by inefficiency and the potential for inaccuracies, often leading to proxy attendance. The main aim of this research work is to manage and monitor the student's attendance by using face recognition technology. This proposed model is mainly categorized four major modules. First module is database creation. Second module is face detection. Then third module is face recognition and final module is automatic attendance updating process. Student images are compiled to create a comprehensive database, ensuring inclusivity across the class roster. The system utilizes the face recognition library, which relies on deep learning based algorithms for face detection and recognition during testing. This face recognition part Convolutional Neural Network algorithm is used. The system matches detected faces with the known database and marks attendance, ensuring a streamlined and accurate attendance tracking process. This innovative approach has the potential to revolutionize attendance management in educational settings, offering a contactless and efficient solution while mitigating proxy attendance concerns. The proposed model is to compare the accuracy level of face recognition. 2023 IEEE. -
Vehicular Propagation Velocity Forecasting Using Open CV
This work presents a predictive learning driven methodology for recognizing the vehicular velocity. The developed model uses machine vision models to trace and detect vehicular movement in timely manner. It further deploys a machine tested framework for estimation of its velocity on basis of the accumulated information. The technique depends upon a CNN model that is validated with a standardized instances of vehicular scans and corresponding velocity parameters. The proposed model generates good efficiency and robustness in determining velocities across test conditions which encompass various kinds of vehicles and lighting scenarios. An optimal vehicular frequency is noted with heavy-weight vehicles in place in comparison to other vehicles. A mean latency period of 1.25 seconds and an error rate of 0.05 is observed with less road traffic in place. The suggested approach can be of great help in transportation systems, traffic monitoring and enhancing road safety. 2023 IEEE. -
Machine Learning Based Recession Prediction Analysis Using Gross Domestic Product (GDP)
This research article aims to explore the prediction and analysis of recessions, with a particular focus on Gross Domestic Product (GDP). The study examines the impact of recessions on different countries, namely India, USA, Germany, China, and Bangladesh, while also considering the influence of the COVID-19 pandemic on these nations in relation to the recessionary effects. Furthermore, the study lists many machine learning techniques that could be used to anticipate recessions. This research mainly focuses on predicting recession using different machine learning models. The research not only provides an in-depth analysis of the recessionary impacts on different economies but also serves as a foundation for future implementation of these algorithms for accurate recession prediction and proactive economic decision-making. This research study mainly focuses on machine learning algorithms like Random Forest, Support Vector Machines and Regression Model. The GDP prediction comparison is taking last twenty years data. This is mainly compared before and after COVID-19 situation. 2023 IEEE. -
Identification of Cyberbullying and Finding Target User's Intention on Public Forums
Numerous cybercriminals are active in the online realm, carrying out cyber-crimes according to predetermined and preplanned agendas. Cyberbullying, which was formerly limited to physical limits, has now expanded online as a result of technology advancements. One type of cyberbullying is denigration or insult. The cyberbullying cases are in exponential rise in social media as per the reports of Computer Emergency Team by Sri Lanka. Insulting words are changeable in dynamic and the same terminology may have numerous meanings depending on the context. Bullying cannot be defined just because a statement comprises such a term. As a result, when classifying comments, standard keyword detecting approaches are insufficient. Other languages also may have dealt with this issue by utilizing lexical databases like WordNet, which might give synonyms as well as homonyms for words. Because no adequate lexical database mainly for the English language has been built, recognizing a word like bullying is difficult. As a result, employed rules to solve the problem. Facebook comments containing profanity were gathered, outliers were eliminated, and the remaining messages were pre-processed. Five feature extraction rules were employed to assess insult in the text. Following that, used the Support Vector Machine (SVM) technique. Using an F1-score of 85%, the findings demonstrate that when compared to existing works, SVM performs better. The focus on English language cyberbully identification, which has never been addressed earlier, distinguishes this study. 2023 IEEE. -
Predictive Analytics for Stock Market Trends using Machine Learning
Navigating the intricacies of stock market trends demands a novel approach capable of deciphering the web of financial data and market sentiment. This research embarks on a transformative journey into the realm of machine learning, where we harness the power of data to forecast stock market trends with increased precision and accuracy. Commencing with an exploration of stock market dynamics and the inherent limitations of traditional forecasting techniques, this paper takes a bold step into the future by embracing the potential of machine learning. The study begins with an in-depth analysis of data preprocessing, unraveling the complexity of feature selection and engineering, setting the stage for a data-driven odyssey. As our exploration progresses, we dive into the deployment of diverse machine learning algorithms, including linear regression, decision trees, random forests, and the formidable deep learning models such as recurrent neural networks (RNNs) and long short-term memory networks (LSTMs). These algorithms act as our guiding lights, revealing intricate patterns concealed within historical stock price data. Our journey reaches new heights as we recognize the significance of augmenting predictive models with external data sources. Incorporating elements like news sentiment analysis and macroeconomic indicators enriches our understanding of the market landscape, enhancing the predictive capabilities of our models. We also delve into the crucial aspects of model evaluation, guarding against overfitting, and selecting appropriate performance metrics to ensure robust and reliable predictions. The research reaches its zenith with a meticulous analysis of real-world case studies, providing a comparative perspective between machine learning models and traditional forecasting methods. The results underscore the remarkable potential of machine learning in predicting stock market trends more accurately. 2023 IEEE. -
Impact of Expert Academic Teaching Quality and its Performance Based on BiLSTM-Deep CNN Network
Undergraduate and postgraduate students from eight different departments at a UK institution participated in organized conversations about the impact of teachers' research activities on their education. In both samples, positive responses greatly outnumbered negative ones. There was an increase in positive feedback on professors' research when the overall quantity and quality of research in a specific field (as measured by Research Assessment Exercise [RAE] ratings) improved. Undergraduate samples with higher RAE scores were more likely to have negative feedback on research than graduate student samples. Both graduate and undergraduate students agreed that lecturers' research increased the instructor's credibility, relevance, and knowledge, as well as piqued and maintained their own interest, engagement, and drive. Data processing, feature selection, and model training are the first steps in the proposed approach. The data are changed from their raw form into a form suitable for academic use during the data pre-processing phase. They are employing Information Gain and Symmetric Uncertainty for feature selection. Following the feature selection process, the models are trained using BiLSTM-CNN. Both the BiLSTM and the CNN methods are inferior to the proposed method. 2023 IEEE. -
A Comprehensive Review on Image Restoration Methods due to Salt and Pepper Noise
Digital images are well-use in various fields like satellite communication, mobile communication, medical and security. Visualized information helps the people to understand the things easily by seen. Improper capturing, age of camera lens, imperfect storage and transmission leads to introduce noise in the image. Gaussian noise, salt and pepper/impulse noise and speckle noise may affect the original image due to aforementioned reasons. Out of these, impulse noise/salt and pepper noise is one of the major types, degrades the image with black and white spots it results loss of required information. Hence, restoration of ground- truth image from such type of noisy image is a challenging task to provide quality and clarity visuals to users. Several linear and non-linear methods have been proposed by researchers since more than four decades. Nonlinear methods based on; median filtering approach; adaptive median filter approach; median filter with switching condition; and median filter with rank order type; are proposed from early 1980s onwards. All of these operated directly on pixels in spatial domain. Hence, they are very easy to implement and most of them are not that much robust at middle and higher noise density circumstances. Further, various researchers have been implemented linear methods such as wavelet transform methods like SWT and DWT. Majority of these are works well upto 50% noise density conditions and very few works well on higher and multiple noise density conditions also. To overcome these problems CNNs based methods have been developed tremendously by various researchers from last decade and these methods require huge database to train the network model. Most of these, achieved good accuracy rates at higher and multiple noise conditions. Hence, here a detailed review report is presented on impulse noise removal methods with their Peak Signal to Noise Ratio (PSNR). 2023 IEEE. -
Impact of Machine Intelligence on Clinical Disease Outbreak Prediction
This research paper examines the utilization of Artificial Intelligence (AI) in disease outbreak prediction and its importance in public health. It explores the hurdles associated with predicting disease outbreaks, including data quality and accessibility, ethical considerations, algorithmic bias, and integration and interpretability challenges. The paper presents an overview of AI techniques applied in healthcare and their relevance to forecasting disease outbreaks. Case studies demonstrate the efficacy of AI -based models in predicting infectious diseases, vector-borne diseases, and epidemics/pandemics, employing diverse data sources. The limitations and future prospects of AI in disease outbreak prediction are addressed, accompanied by recommendations for enhancement. In conclusion, the paper highlights AI's potential to revolutionize disease outbreak prediction, leading to proactive public health interventions and improved response strategies. 2023 IEEE. -
A Voting Enabled Predictive Approach for Hate Speech Detection
In today's digital environment, hate speech, which is defined as disparaging and discriminating communication based on personal characteristics, presents a big difficulty. Hate crimes and the rising amount of such content on social media platforms are two examples of how it is having an impact. Large volumes of textual data require manual analysis and categorization, which is tedious and subject to prejudice. Machine learning (ML) technologies have the ability to automate hate speech identification with increased objectivity and accuracy in order to overcome these constraints. This article intends to give a comparative analysis of various ML models for the identification of hate speech. The proliferation of such content online and its negative repercussions on people and society are explored, as is the necessity for automated hate speech recognition. This paper intends to support the creation of efficient hate speech detection systems by performing a comparative analysis of ML models. Random forest records the best performance with higher accuracy and low response delay period for hate speech detection. The results will help enhance automated text classification algorithms and, in the end, promote a safer and more welcoming online environment by illuminating the benefits and drawbacks of various approaches. 2023 IEEE. -
AI Sovereignty in Autonomous Driving: Exploring Needs and Possibilities for Overcoming Challenges
With the development of artificial intelligence, advancements in navigation systems for self-driving cars have become a new direction over the last decade. The inclusion of AI-driven actuators in autonomous vehicles has broken the barriers in terms of real-time high-quality data processing resources, accuracy of decisive actions and generalization of environment-action pairs. Upgradation from a car with no automation to a car with minimal to no human intervention has become a boon of AI, as it resolves most of the transportation problems on roads, including human error, lack of visibility in adverse weather conditions, tiredness of drivers in long journeys, etc. This study focuses on AI-enabled tasks, including object detection and identification, lane detection, notification for lane departure and reinforcement learning from the operational environment. However, there exist serious issues in deploying AI-empowered modules in autonomous cars, as the consumer rights to explain, trustworthiness, and reliability of the machine have not yet met the requirements. Our work explores the needs and prospects of AI sovereignty in autonomous driving by overcoming the aforementioned issues so that the healthy progress of technological society can take care of the future world. 2023 IEEE. -
Prediction of Rainfall Using Seasonal Auto Regressive Integrated Moving Average and Transductive Long Short-Term Model
One of the most crucial parts of the practical application in recent years has been the analysis of time series data for forecasting. Because of the extreme climate variations, it is now harder than ever to estimate rainfall accurately. It is possible to forecast rainfall using a number of time series models that uncover hidden patterns in past meteorological data. Choosing the right Time Series Analysis Models for predicting is a challenging task. This study suggests using a Seasonal Auto Regressive Integrated Moving Average (SARIMA) to forecast values that are similar to historical values that exhibit seasonal patterns. Twelve years of historical weather data for the city of Lahore (from 2005 to 2017) and Blora Regency are taken into account for the prediction. The dataset underwent pre-processing operations like cleaning and normalisation before to the classification procedure. For classification, Transductive Long Short-Term Model (TLSTM) is employed which has learned the dependency values where the memory blocks are recurring and capable of learning long-term dependencies on this model. Further, TLSTM's goal is to increase accuracy close to the test point, where test points are selected as a validation group. The performance of the models has been assessed based on accuracy (99%), precision (98%), recall (96%) and fl-score (98%). Proposed SARIMA model showed optimistic results when compared to existing models. 2023 IEEE. -
Augmented Reality Based Medical Education
The education in medical field requires both theoretical knowledge and practical knowledge. It is important for medical student to acquire effective practical skills. Since the students apply the theoretical knowledge in practical manner in human body. Human body is very volatile, gentle, and difficult system. If a student apply trial in the humans for practical knowledge, there may cause the human error which leads to death of the person. To avoid this, the proposed system 'Augmented Reality Based Medical Education' is useful. Augmented reality makes the learning process more interactive and interesting. It can reproduce specific circumstances that assist students to rehearse with virtual objects that look like the human body and organ. Like traditional learning, it does not require real patients. By this way, augmented reality prevents risk of human life. Medical education with augmented reality extensively provides real time experiences. It has low risks and also affordable. When any human error occurs, there is no human loss. So the human life can be prevented by the system. The proposed system is developed using tools like Unity which is the complete platform for the developing our application, Vuforia-developer portal, a tool to create image target and Blender which is used to create 3D objects. 2023 IEEE. -
5G Technology Empowering Wireless Technology
Wireless Communication is the means of transferring data from one point to another without the use of any wired means. With reference to wireless communication, wireless sensor Networks (WSN) have also developed in recent times. It can be referred as an infrastructure-less system of wireless devices which can gather and exchange information with the help of a wireless link. The information which is gathered is sent respectively to the base stations and sinks for further developments. Recently, the 5G generation network, the latest Wireless Communication Network operates at a higher frequency range than its predecessor. In this paper, a detailed analysis on the 5G generation cellular network, which is expected to be a key instrument of wireless technologies in the near future is outlined. Also a comparative analysis of different kinds of networks in context to wireless scenario is discussed. It was found that 5G provides the best outcome in terms of high speed and network spectrum bandwidth. 2023 IEEE. -
An Efficient Approach for Obstacle Avoidance and Navigation in Robots
Reinforcement learning has emerged as a prominent technique for enhancing robot obstacle avoidance capabilities in recent years. This research provides a comprehensive overview of reinforcement learning methods, focusing on Bayesian, static, dynamic policy, Deep Q-Learning (DQN) and extended dynamic policy algorithms. In the context of robot obstacle avoidance, these algorithms enable an agent to interact with its physical environment, learns effective operating strategies, and optimize actions to maximize a reward signal. The environment typically consists of a physical space that the robot must navigate without encountering obstacles. The reward signal serves as an objective measure of the robot's performance towards accomplishing specific goals, such as reaching designated positions or completing tasks. Furthermore, successful obstacle avoidance strategies acquired in simulation environments can be seamlessly transferred to real-world scenarios. The promising results achieved thus far indicate the potential of reinforcement learning as a powerful tool for enhancing robot obstacle avoidance. This research concludes with insights into the future prospects of reward learning, high-lighting its ongoing importance in the development of intelligent robotics systems. The proposed algorithm DQN outperforms well among all the other algorithms with an accuracy of 81%, Through this research, we aim to provide valuable insights and directions for further advancements in the field of robot obstacle avoidance using reinforcement learning techniques. 2023 IEEE. -
A Novel Energy-Efficient Hybrid Optimization Algorithm for Load Balancing in Cloud Computing
In the field of Cloud Computing (CC), load balancing is a method applied to distribute workloads and computing resources appropriately. It enables organizations to effectively manage the needs of their applications or workloads by spreading resources across numerous PCs, networks, or servers. This research paper offers a unique load balancing method named FFBSO, which combines Firefly algorithm (FF) which reduces the search space and Bird Swarm Optimization (BSO). BSO takes inspiration from the collective behavior of birds, exhibiting tasks as birds and VMs as destination food patches. In the cloud environment, tasks are regarded as autonomous and non-preemptive. On the other hand, the BSO algorithm maps tasks onto suitable VMs by identifying the possible best positions. Simulation findings reveal that the FFBSO algorithm beat other approaches, obtaining the lowest average reaction time of 13ms, maximum resource usage of 99%, all while attaining a makespan of 35s. 2023 IEEE. -
TSM: A Cloud Computing Task Scheduling Model
Cloud offers online-based runtime computing services through virtualized resources, ensuring scalability and efficient resource utilization on demand. Resource allocation in the dynamic cloud environment poses challenges for providers due to fluctuating user demand and resource availability. Cloud service providers must dynamically and economically allocate substantial resources among dispersed users worldwide. Users, in turn, expect reliable and cost-effective computing services, requiring the establishment of Service Level Agreements (SLAs). Resource distribution uncertainty arises in view of the dynamicity of the cloud, where VMs, memory capacity requirement, processing power, and networking are allocated to user applications using virtualization technology. Resource allocation strategies must address issues such as insufficient provisioning, scarcity, competition, resources fragmentation. CPU scheduler plays a crucial role in task completion, by selecting job from queue considering specific requirements. The Task Scheduling Model (TSM) algorithm improves scheduling by considering expected execution time, standard deviation, and resource completion time, aiming to address resource imbalances and task waiting times. The research discusses previous work, presents experimental findings, describes the experimental setup and results, and concludes with future research directions. 2023 IEEE.