2023

Browse

Now showing 1 - 20 of 52

Efficient Ensemble-Based Approaches to Personal Health Mention Detection
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-04-30) Nower, Nuzhat; Kamal, Fida; Khan, Alvi Aveen
An important component of public health surveillance is the analysis of personal health-related posts on social media platforms, known as Personal Health Mention (PHM) Detection. PHM detection is essential to quickly detecting epidemics, allowing health organisations to prepare themselves and warn the general public to take precautionary steps. One of the key complexities of this task is the informal nature of the language used in social media, which makes it difficult to understand their context. The architectures that are capable of discerning context are also computationally expensive to use and often mistrusted in the medical community due to their decision-making strategies being hidden behind a black box. In this thesis, we address each of these issues separately. We introduce four transformer-based ensemble architectures that have not been previously explored in the PHM domain and show that these architectures can achieve state-of-the art results across the domain. Combined with computationally efficient training mechanisms, our architectures also use fewer resources than existing ones. Addi tionally, we provide methods to explain the outputs produced by the architectures in order to address the concerns related to explainability.
Attack and Anomaly Detection in IoT Devices using Federated Learning
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-04-30) Adib, Mosabbir Sadman; Raf, Moshiur; Pranto, MD Jabear Hossain
There has been a lot of focus from governments, universities, and businesses in recent years on the intersection of cybersecurity and machine learning (ML) for the Internet of Things (IoT). The Internet of Things (IoT) can be made more secure and efficient in the future through the groundbreaking concept of federated cybersecurity (FC). This new idea has the ability to efficiently identify security problems, implement countermeasures, and contain them within the IoT network infrastructure. Cybersecurity goals are met through the federation of a shared and learned model among several actors. Protecting the insecure IoT environment requires privacy-aware ML models like federated learning (FL).
PPoS: An Optimized Consensus Protocol for IoT Devices
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Anan, Tasnim Ferdous; Mahi, Abdullah Ibne Masud; Arnob, Tausif Khan
Internet of Things (IoT) devices are currently on the rise, with homes and businesses constantly adding and replacing old mechanisms with new, smart IoT devices. How ever, security vulnerabilities are also becoming more apparent as a result and the use of Blockchains can help reduce some aspects of these vulnerabilities. Most of these blockchains are based on private, permissioned architectures, which while are perfect for private solutions, present a new set of problems when applied to public infrastructures. We thus tried to focus on the applications of public blockchains on IoT devices. TinyEVM presents a novel solution to this problem. However TinyEVM is optimized mostly for one-to-one communication, thus having inherent scalability issues. Current consensus protocols for side-chains can be used to solve this problem, but they are either not optimal or have security risks. In this paper we thus propose a novel consensus pro tocol for side-chains, tailored for use in IoT devices in conjunction with TinyEVM which would enable the use of the public blockchain, Ethereum on most IoT devices. Periodic Proof of Stake (PPoS) is a consensus protocol based on Delegated Proof of Stake (DPoS) but with a Trust model and a focus on minimizing network traffic through performing consensus periodically and multi-casting transactions to selected nodes instead of whis pering. PPoS also provides flexibility to choose between security and energy-efficiency. Traffic, performance and security analysis shows the improvements compared to existing consensus protocols.
Underwater Image Enhancement based on Residual and Adversarial Network
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Rahman, Md Tosadduk; Tanha, Md. Tawratur Rashid; Hossain, Ishrak
Due to the complexity and constraints of the underwater environment, underwa ter picture enhancement is a difficult task. In order to improve underwater images that have problems with low contrast, blurriness, and color mistakes, this research suggests a deep learning-based technique. Residual Networks (ResNet) and Super Resolution Generative Adversarial Networks (SRGANs) are combined in the sug gested method. In order to restore fine details and improve overall contrast and sharpness, ResNet extracts residual information. SRGANs produce enhanced under water picture versions at high resolution, enhancing visual integrity. Extensive testing on several underwater picture datasets reveals the suggested method’s superior performance. Comparing it to cutting-edge methods, objective quality indicators such as contrast augmentation, image sharpness, and color accu racy confirm its efficiency. Qualitative evaluations show that the underwater pho tographs have significantly improved in terms of contrast, blurriness, and color re production. This increases their ability to be analyzed and interpreted as well as their visual appeal. Marine research, underwater robots, and inspection systems can all benefit from better underwater image quality. Improved visual quality is bene ficial for accurate underwater object identification, biodiversity measurement, and extending our understanding of underwater ecosystems. In conclusion, this study provides a deep learning-based technique for enhancing underwater image quality that combines ResNet and SRGANs. The method addresses low contrast, blurriness, and color mistakes to produce notable improvements. Its effectiveness is supported by the experimental findings and qualitative evaluations, emphasizing its potential to advance underwater photography methods and applications
Vulnerability Analysis of WebAssembly Binaries
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Saif, Farhan; Sikder, Shihab; Kabeer, Adib Abrar
The evolution of web technologies has brought forth innovative coding structures, among which JavaScript and WebAssembly stand out prominently. This paper presents Wasmosys, a state-of-the-art source code analyzer designed to generate and unify Abstract Syntax Trees (ASTs) for JavaScript and WebAssembly code. It aims to pave the way towards advanced vulnerability detection and mitiga tion in these modern web environments.Wasmosys tackles two major challenges: creating a seamless combination of separate ASTs and standardizing AST labels for JavaScript and WebAssembly. The system comprises four primary modules. The first two modules, written in JavaScript and C respectively, generate ASTs from JavaScript source files and WebAssembly Text (WAT) files. The third mod ule constructs a unified AST from the generated JS and Wasm ASTs, and the fourth module, a connector written in python, links the system with a Neo4j graph database hosted in a Docker container.Despite its capabilities, tested on a limited version of WasmBech, Wasmosys currently presents certain limitations, including the use of AST over Code Property Graphs (CPG), manual AST uni fication, and constraints in the experimental dataset. These limitations serve as insights for future development, hinting at the prospect of an even more robust and accurate tool for JavaScript and WebAssembly code analysis
Preventing Data Loss using Raft Consensus Algorithm in a Decentralized Database System
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Hafiz, Md. Muhtaseen; Zaman, A.K.M Nafiz; Shaf, Md Shadman
In today’s digital era, decentralized database management systems have gained significant attention due to their ability to provide scalability, fault tolerance, and improved performance. However, ensuring data integrity, preventing data loss, and maintaining data consistency in such systems remain challenging tasks. This thesis addresses these challenges by proposing a peer-to-peer gossip-based solution that leverages the Raft consensus algorithm and replicated log method. The proposed solution focuses on making each node in the database cluster a witness to transactions, allowing for consensus on the current state of the database. By utilizing gossip-based protocols, transaction information is disseminated among nodes, ensuring that updates reach all relevant participants. The Raft consensus algorithm is employed to achieve agreement on the committed transactions, while the replicated log method synchronizes transaction logs across all nodes. The objectives of this thesis include preventing data loss, maintaining data con sistency, and meeting high transaction and view request targets. With a target transaction rate of 1000 transactions per second and a target view request rate of 10000 requests per second, the solution aims to deliver robust performance and reliability. By combining the peer-to-peer gossip-based approach, Raft consensus algorithm, and replicated log method, the proposed solution offers benefits such as fault tolerance, scalability, and data consistency. The thesis contributes to the field by addressing the limitations of current database systems and proposing an innovative solution that ensures data integrity in de centralized environments. The limitations and complexities of Direct Mail, Anti Entropy, and Rumor Mongering techniques are analyzed, leading to the devel opment of a more effective and efficient solution. The solution’s architecture, mechanisms, and protocols are designed to meet the specified targets and provide a reliable foundation for decentralized database management systems. Through simulations and performance evaluations, the proposed solution demon strates its effectiveness in preventing data loss, maintaining data consistency, and meeting the specified transaction and view request targets. The results highlight the solution’s scalability, fault tolerance, and ability to handle high transaction rates. In conclusion, this thesis presents a peer-to-peer gossip-based solution that lever ages the Raft consensus algorithm and replicated log method to prevent data loss and ensure data consistency in decentralized database management systems. The solution offers a robust and scalable approach, addressing the limitations of exist ing techniques. With its potential applications in various domains, the proposed solution contributes to the advancement of decentralized database management systems, providing a foundation for reliable and high-performance data storage and processing.
Real time Gaze Tracking in Remote Proctoring - A Study of Appearance-based Gaze Estimation
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Onim, Nafiul; Shahid, Mirza Sadaf; Quayes, Muhammad Rafsan
Our thesis aims to address the critical issue of academic dishonesty in online examinations by proposing a proctoring system that integrates eye gaze tracking technology for the detection of suspicious behavior. The study begins by discussing the existing challenges of current examination systems and identifying the problems that need to be addressed. It emphasizes the necessity for a more advanced proctoring system with gaze tracking capabilities to effectively deter attempts at academic dishonesty. The research is divided into two main parts: the selection of an appropriate model and the incorporation of proctoring functionalities. Two models were chosen for evaluation, namely iTracker, which was pre-trained on the GazeCapture dataset, and L2cs-net, which we trained on the MPIIFaceGaze dataset. The findings from these experiments indicate that L2cs-net outperforms iTracker in terms of accuracy, speed, and latency but only when supplied with the processing power of a GPU, without one iTracker is better. Regarding the proctoring system aspect, it is noted that most of the existing research is commercially driven, with limited academic contributions. To optimize the proctoring system for online exams, we recognize the significant value of examinees’ eye gaze and define important regions on and off the screen through calibration using “magic pixels”. Moreover, we attribute cheating criteria using a formulated equation that takes into account factors such as Count, Frequency, Duration, and Regression. Two potential approaches for the proctoring system, namely Thresholding and Machine Learning (ML), are considered. However, our focus lies on the development of a thresholding-based approach. Overall, this thesis presents a comprehensive exploration of academic dishonesty in online examinations, proposes a proctoring system using eye gaze tracking technology, and compares the performance of different models and methodologies. The findings contribute to the advancement of proctoring systems and provide insights for the development of more effective measures against academic dishonesty.
Answer-agnostic Bangla Question-answer Pair Generation Using Transformer-based Approaches
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Altaf, Md Sajid; Ekram, Syed Mohammed Sartaj; Rahman, Adham Arik
High-resource languages, such as English, have access to a plethora of datasets with various question-answer types resembling real-world reading comprehension. However, there is a severe lack of diverse and comprehensive question-answering datasets in under-resourced languages like Bangla. The ones available are either translated versions of English datasets with a niche answer format or created by human annotations focusing on a specific domain, question type, or answer type. To address these limitations, we introduce BanglaRQA, a reading comprehension-based Bangla question-answering dataset with various question answer types. BanglaRQA consists of 3,000 context passages and 14,889 question-answer pairs created from those passages. The dataset comprises answerable and unanswerable questions covering four unique categories of questions and three types of answers. In addition, we also implemented four different Transformer models for question-answering on the proposed dataset. The best-performing model achieved an overall 62.42% EM and 78.11% F1 score. However, detailed analyses showed that the performance varies across question-answer types, leaving room for substantial improvement of the model performance. Furthermore, we demonstrated the effectiveness of BanglaRQA as a training resource by showing strong results on the bn_squad dataset. We focus on Bangla Question-answer pair generation for the next part of our work. Bangla, being a less explored language in NLP, lacks comprehensive research in the do main of question-answer pair generation. We focus on developing this untapped sector by fine-tuning BanglaT5, a generative model on the BanglaRQA dataset. The quality of the generated questions is first evaluated using various metrics. The best-performing model, BanglaT5, achieved a BLEU score of 21.56 and a BERT score of 85.04, indicating that the generated questions exhibit decent quality. Subsequently, the research progresses toward the main task of generating question-answer pairs. The quality of the generated pairs is evaluated through human assessment and baseline comparison, demonstrating that the generated QA pairs possess comparable quality to human-annotated QA pairs. Therefore, this work proposes an end-to-end Question-Answer-Generation (QAG) pipeline and presents a reading-comprehension-based dataset, that has the potential to contribute to future research
A peer to peer blockchain based approach for blood donation community
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Kamal, Minhaz; Abdullah, Chowdhury Mohammad; Shaiara, Fairuz
Medical Image Synthesis using Generative Adversarial Network
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Risha, Antara; Islam, Shaira Saiyara; Tahsin, Anika
Medical image synthesis has emerged as a promising technique in the field of healthcare, enabling the generation of realistic medical images for various applica tions. This study focuses on medical image synthesis using Generative Adversarial Networks (GANs) applied to the IDRID dataset, which contains retinal images for diabetic retinopathy analysis. The objective of this research is to explore the potential of GANs in generating synthetic retinal images that closely resemble real patient data. The IDRID dataset provides a valuable resource for training and evaluating the GAN model. By leveraging the power of GANs, the proposed framework aims to generate high-quality synthetic retinal images with similar char acteristics and visual appearance to real patient images. This has the potential to augment the existing dataset, expand its diversity, and improve the performance of diagnostic and treatment algorithms. The methodology involves training a GAN architecture consisting of a generator and a discriminator network. The generator network learns to generate synthetic retinal images from random noise, while the discriminator network evaluates the authenticity of the generated images. The two networks engage in an adversarial training process, where the generator aims to fool the discriminator into classifying the synthetic images as real. Evaluation of the synthesized retinal images includes quantitative metrics such as structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and analysis to as sess the similarity and quality of the generated images compared to real IDRID dataset images. The outcomes of this research provide insights into the capabili ties of GANs in generating realistic retinal images from the IDRID dataset. The generated images have the potential to enhance the limited availability of labeled medical data, facilitate algorithm development, and support computer-aided di agnosis systems. The findings contribute to the broader field of medical image synthesis, showcasing the potential of GANs in improving healthcare outcomes through enhanced image data availability and diversity
Deep Learning Approach: Image Captioning in French and Arabic Language
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Keita, Abdoulaye; Hamadou, Mohaman Dairou; Asag, Mazen Abdulwahab Mahyoub Salem
This research report introduces a novel dataset of French captions translated from the Flickr30k dataset using different translation models, namely we have Google Trans late and the powerful Transformers: T5 Small and T5 base models. A novel dataset of French captions means creating fresh data collection by translating existing captions from the Flickr30k dataset into French. The Flickr30k dataset is valuable for training and evaluating image captioning models in French. The main objective is to address the problem of generating precise image captions in French. The performance of an image captioning model is evaluated on the translated datasets, employing ResNet-50 for image feature encoding and LSTM network with attention in generating captions. These results demonstrate that the accuracy of image captions varies depending on the translation(or Language) models, with the Trans formers models outperforming Google Translate. The proposed approach achieves state-of-the-art performance in generating accurate French captions when combined with ResNet-50 and LSTM network with attention. The findings contribute to the field of image captioning and machine translation for French speakers, highlighting the importance of using advanced translation models for improved caption accuracy and other NLP tasks in French. Furthermore, this research provides insights into the potential of smaller-scale models in limited data scenarios. Based on our findings, we can explore alternative translation models, and data aug mentation techniques, and consider multi-modal approaches that could lead to more accurate and contextually relevant captions and the potential of this approach in other languages
Damaged Building Detection Using Global Local Attention
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Khadija, Nejd
Dark Triad detection and analysis from social media text
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Morsalina; Fairoz, Fariha; Anjum, Fariha
The increased use of social media platform usage such as Facebook has given an oppor tunity to express one’s thoughts and ideas to everyone. Social media posts can be used as a medium for determining different psychological traits, such as dark triad character istics. In our thesis, we used peoples’ social media posts, and using those posts we tried to detect presence of dark triad traits based on the hand crafted features extracted from their posts. We also have shown, the usage of code mixing as a effective feature. We have used traditional machine learning models, ensemble models as well as transformer based language models to find the best possible outcome. For finding linguistic fea tures analysis we have used Polarity of text, subjectivity, lexical density, word tokens, word count, avg word length, word freq dist, stopword count, part of speech, Topic seg mentation and many more. We then compared the performance of all the models.For all the traits, ensemble of 4 traditional machine learning models outperformed all other models.The highest accuracy achieved in narcissist detection was 96.36% , for Machi avelli it was 98.27%, and for psychopathy it was 99.62%.
An Ensemble Method for Cancer Classification and Identification of Cancer-Specific Genes from Genomic Data
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Rizwan, Siana; Tabassum, Farzana; Islam, Sabrina
Classifying cancer using gene expression can be an important tool for under standing the specific characteristics of a patient’s cancer and for guiding the most appropriate treatment approach. By identifying the specific genes that are involved in the development and progression of a particular cancer, it may be possible to tailor treatment to target those genes and improve outcomes for the patient. In addition, by understanding the genetic makeup of a patient’s cancer, it may be possible to identify clinical trials or targeted therapies that may be more effective for that patient. Here, in our study, we worked with the TCGA Pan Cancer dataset where we used the RNA-seq data for analyzing the gene expres sions. The dataset comprises 33 types of cancer. Our study mainly focuses on implementing an explainable AI-based panCancer classification approach using gene expression analysis. The goal is to accurately detect the type of cancer in in dividuals within a short time. We employed seven classifier algorithms- Logistic Regression, SVM, XGBoost, Random Forest, MLP, 1-D CNN, and TabNet. To enhance the performance of the models, we utilized feature selection techniques such as Lasso, SelectFromModel, Select-K-Best, and ElasticNet. SelectFrom Model with 500 features yielded the best performance. We applied ensemble methods of probability averaging and max voting, with probability averaging achieving the highest accuracy of 96.60%. Validation of the selected features’ contribution and comparison with gene sets from DESeq2 analysis confirmed their significance and relevance. This approach provides insights into cancer specific molecular mechanisms and pathways. Overall, our study demonstrates the effectiveness of feature selection in reducing dimensionality while maintain ing predictive power and biological relevance
Seamless Service Migration in Cloud Edge Networks with QUIC
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Amio, Zibran Zarif; Choudhury, Fida Waseque; Mohaimen, Mohammed
Cloud computing is hoped to replace traditional computing paradigms in the near future, as the Internet becomes a more integral part of our lives, more and more computing resources are being hosted in the cloud. One of the common techniques used by cloud service providers is to migrate cloud-based applications from one server to another for a variety of reasons. This thesis aims to add on the possible strategies of container migration in the cloud using QUIC in an innovative way. The idea is to use a dual-path extension of QUIC to ensure that the user’s Quality of Experience is not hampered by the migration of the application hosted in the cloud server. This approach is coined as Dual-path in our thesis. Cloud services are provided via containers that are processes running inside of the servers. Due to a number of conditions such as load balancing, resource balancing, hardware failure or maintenance etc. the container has to be migrated from one server to another. Traditional live migration techniques like Pre-Copy and Post-Copy con sists of three rudimentary phases: iterative push phase, pulling of faulted pages and stop-and-copy (control transfer). During the control transfer phase the cloud service is unavailable and suspended until the container state is fully replicated to another tar get server. This introduces a downtime, hampering the end user’s quality of experience. Furthermore, pulling faulted pages involves performance degradation which is not desir able. To mitigate the limitations identified in the traditional live migration techniques, we formulate the dual-path migration scheme. Dual-path migration is an endeavor to redefine live migration techniques where an end user can simultaneously be connected to two servers at any given time. In this approach, once the migration is triggered the end user is dually connected to both the servers capable of requesting and receiving service from any of them. Initially, service is provided to the end user from the source server (traditional single path). Once the migration is triggered the container in the source server does not get suspended like the traditional schemes. Rather it will keep providing service to the end user and the transfer of control will be executed in the background. During this control transfer the end user can request data from any of the two servers. Since the end user is concurrently connected to both servers, the server having the requested data can respond. Once the background migration is completed it will simply terminate connection with the initial server and switch to the target server (again single path). The key attainment in this approach is its negligible downtime and performance upgrade. It also solves synchronization issues between the servers. In this work, we compare and contrast between traditional live migration techniques and our proposed dual-path migration by mathematically analysing post-copy migration using QUIC and dual-path migration, we show that under certain circumstances the dualpath migration scheme performs better than post-copy migration scheme.
Improving Zero-Shot Semantic Segmentation using Dynamic Kernels
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Tajwar, Tauseef; Rahman, Muftiqur; Chowdhury, Taukir Azam
Zero-shot Semantic Segmentation (ZS3) is a daunting task since it requires segmenting items into classes that were never seen during training. One popular method is to divide ZS3 into two sub-tasks: creating mask suggestions and assign ing class labels to individual pixels inside those regions. However, many existing approaches have difficulty producing masks with sufficient generalization capa bilities, resulting in notable performance constraints, particularly on unknown classes. In this regard, we propose using “Dynamic Kernels” to improve object understanding within a ZS3 model during the training phase. We want to pro duce superior mask suggestions that permit a more accurate representation of the objects by harnessing the intrinsic inductive biases of these kernels. These specialized agents, known as dynamic kernels, adjust based on data taken from visible classes, allowing them to obtain insights on unseen things. In addition, for segment classification, our proposed system utilizes the Contrastive Language Image Pre-Training (CLIP) architecture. This integration improves the model’s generalizability by utilizing its cross-modal training capabilities. The utilization of dynamic kernels in conjunction with CLIP proves to be advantageous as it allows for finer granularity in processing, enabling performance enhancements for both seen and unseen classes. Our proposed ZSK-Net surpasses the existing state-of-the-art methods by achieving a remarkable improvement of +10.4 and +0.9 in hIoU on the Pascal VOC and COCO-Stuff datasets, respectively.
Efficient Leader Head Selection In Vanet By Minimizing Message Overhead
(2023-05-30) Aboubakar, Mouhammad; Mamoudou, Abdoulaye; Nom, Aminata
Vehicular Ad-hoc Networks (VANETs) require efficient leader head selection algorithms to optimize network management tasks and communication overhead In this paper, we offer a proactive approach technique to reduce message overhead while retaining effective network performance for leader head selection in VANETs. To create clusters and choose leader heads, the system makes use of predictive models of vehicle motion and communication patterns. The proposed technique achieves better scalability, decreased communication costs, and improved network performance by minimizing the number of messages exchanged and optimizing cluster formation. The algorithm's success in decreasing message overhead and ensuring effective leader head selection in VANETs is demonstrated by simulation results. Vehicle-to-vehicle (V2V) communications have rapidly advanced in recent years, opening the door for new applications tackling concerns like autonomous driving, traffic efficiency, and vehicle safety. These applications may greatly enhance the driving experience, travel time, fuel efficiency, traffic safety, and other elements directly connected to automobiles. In many of these circumstances, a coordinator vehicle is necessary to coordinate the right-of-way among many cars. Such a coordinator or leader vehicle is essential in many situations where a cooperative goal is desired for all cars in the group. Effective leader head selection algorithms are crucial in the context of vehicular ad hoc networks (VANETs) for optimizing network management activities and reducing communication overhead. The proactive approach strategy that is suggested in this study attempts to lower message overhead while maintaining efficient network performance in VANETs. To generate clusters and choose leader heads, the method makes use of predictive models of vehicle movements and communication patterns. The suggested method delivers higher scalability, lower communication costs, and increased network performance by utilizing predictive models. This is accomplished by optimizing cluster formation and reducing the volume of communications transferred during the leader-head selection process. To choose the best leader, the system considers a number of variables, including vehicle velocity, proximity, connection, and communication patterns. The algorithm can efficiently divide the leadership position among cars in a way that improves network performance and lowers communication costs by taking these aspects into account. Results from simulations show how the suggested technique is successful in lowering message overhead and guaranteeing efficient leader head selection. The simulations illustrate how the algorithm can build effective clusters and choose appropriate leader heads, improving throughput, latency, and reliability across the network. The algorithm's proactive strategy and implementation of predictive models let it adapt effectively to VANETs' dynamic nature, where cars are continually moving and entering/leaving the network. The development of vehicle-to-vehicle (V2V) communications has created new possibilities for applications relating to autonomous driving, traffic efficiency, and vehicle safety in a more general sense. The existence of a coordinator or leader vehicle is essential in many of these scenarios in order to promote coordination and collaboration among several vehicles. In order to enable smooth coordination and collaboration inside VANETs, the suggested leader head selection algorithm solves this need by effectively recognizing and choosing leaders' heads. Overall, the proactive strategy and technique based on predictive modeling described in this study provide solutions that have promise for lowering message overhead, improving cluster formation, and ensuring efficient leader head selection in VANETs. These developments have the potential to improve a number of directly linked features of vehicles and their interactions in V2V communication scenarios, including the driving experience, journey duration, fuel efficiency, traffic safety, and others.
PPoS: An Optimized Consensus Protocol for IoT Devices
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Anan, Tasnim Ferdous; Mahi, Abdullah Ibne Masud; Arnob, Tausif Khan
Joint Position-based Anomaly Detection using Graph Convolution Network
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Haque, Md. Wasiul; Siddique, Mohammed Afzal; Saju, Md. Hasan
Sleep Stage Classification With Machine learning Models (RandomForestClassifier and DecisionTreeClassifier)
(Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur-1704, Bangladesh, 2023-05-30) Jawo, Musa S; Jasseh, Haddy; Abdifatah, Ismahan
Automated classification of sleep stages is in demand to overcome the limitations of manual sleep stage classification. Analyzing sleep stages manually using neurophysiological signals and inspecting visually is very difficult, time-consuming process. Many techniques have been proposed already in the past decades. Sleep experts, physicians do not have assurance with such techniques concerned with accuracy, specificity and sensitivity. Sleep state classification using electroencephalogram (EEG) signals is crucial for understanding sleep patterns and diagnosing sleep disorders. This thesis aims to improve the accuracy and robustness of sleep state classification by employing a voting technique that combines multiple classification models. The research involves preprocessing and feature extraction from EEG signals, training individual classification models, and applying a voting mechanism to make the final sleep state classification decision. The proposed approach aims to enhance the reliability of sleep stage classification and contribute to the field of sleep medicine. Statistical features are extracted and trained with Decision Tree, Support Vector Machine and Random Forest algorithms with different testing dataset percentage. Results show combination of Random forest and decision tree algorithm achieves 90% of accuracy.