Jahangirnagar University Journal of Electronics and Computer Science

Blockchain-based Secure Healthcare System: A test case of Hyperledger Fabric For Medical Data Management

Mallika Dey Mallika — 2025-06-20

In recent years, the healthcare industry has experienced significant growth in the amount of sensitive information that is being exchanged and stored digitally. This has created new challenges for healthcare organizations in terms of safeguarding patient data from security threats and unauthorized access. Blockchain technology has emerged as a promising solution to address these challenges. This paper proposes a blockchain-based secure healthcare system using Hyperledger Fabric. This system employs smart contracts to manage medical data and access control.

Optimization of Cost Function of LAN Using Particle Swarm Optimization and Genetic Algorithm

Md. Imdadul Islam — 2025-06-20

Under network planning and optimization, the size of service area of a network has to be estimated to minimize the total cost. In this paper we consider the cost function of wired network of telecommunication switching station, which is applied for wired LAN. The function contains both fixed and variable cost with respect to size of the network and does not provide sharp minima on multidimensional plane hence steepest descent method could not solve the cost function. To overcome the situation, we applied Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to acquire the optimum size of the network. The theoretical, PSO and GA provide very close results shown both in table and graph.

Handover Detection of WAN Using Fuzzy Inference System

Neelima Monjusha Preeti — 2025-06-20

In Wide Area Network (WAN) users experience frequent handover while on a high-speed vehicle. In 4G and 5G mobile communication, the coverage area of cells is reduced compared to 3G systems therefore handover occurs frequently. An intelligent handover decision is essential to protect unnecessary handover to reduce network overhead on control plane (the activities of control channels). Moreover, handover failure or forced termination deteriorates the QoS of the network on user plane i.e. the data flow of traffic channels. In this paper, a Fuzzy Inference System (FIS) is developed to make the correct decision of handover. Here four fuzzy variables: velocity of mobile station (MS), SNR of received signal, SINR of received signal and acceleration of MS are used, where each of the variables possess 3 to 5 linguistic values. Around 200 Fuzzy rules are applied to link crisp and fuzzy values. Finally, four defuzzification methods: Centroid, Bisection, MOM and LOM are used to take the decision of handover and a comparison is made among them in both tabular form and using surface plot. All of the four methods ensure correct decision of above 92% and the best one is 100%.

Analysis of ICU data of pediatric heart patients

Sharmin Nahar Sharwardya — 2025-06-20

One of the most fatal conditions for pediatric disease is congenital heart disease (CHD). Reviewing massive databases, comparing them, and mining them for information that can be used to identify, monitor, and treat illnesses like CHD is the key to treating cardiovascular disease. Cardiovascular disease can be predicted, prevented, managed, and treated with great effectiveness using big data analytics, which is well-known all over the world for its useful application in controlling, contrasting, and managing massive datasets. This study analyzes the post-operative ICU data. We analyzed the patient conditions of the ICU patients by using descriptive statistics. Here, we have selected the ICU parameters between different demographic groups by using chi-square test, t test and p value. Besides, this we also used different machine learning method to predict the patient condition. The outcomes will serve as a reference for medical professionals employing big data technology to predict and manage CHD patients in ICU.

Optimizing the Fine-tuning Process of Large Language Models

Mahbub Islam Mahim — 2025-06-20

We present an optimized fine-tuning process for large language models (LLMs) that combines Low-Rank Adaptation (LoRA) and Quantization. Traditional full fine-tuning methods are computationally expensive, requiring significant GPU memory, which limits their accessibility. In our approach, we first quantize the LLaMA-2 7B model and then apply LoRA fine-tuning to that quantized model. We demonstrate that the combination of quantization and LoRA significantly reduces GPU memory requirements while maintaining model performance. Through rigorous experiments, we successfully fine-tuned the 7B LLaMA-2 model using the CodeAlpaca-20k dataset with only 10.8 GB of GPU memory, compared to the 112 GB required by traditional methods. We further developed an inference system using this optimized fine-tuned model for practical deployment.

Study on Transfer Learning Technique for Deepfake Face Detection Using Weighted Average Ensemble Model

Jannatul Mawa — 2025-06-20

Deepfakes represent a significant cybersecurity threat with their ability to create highly convincing fraudulent media. As deepfake technology becomes more sophisticated and accessible, the potential for cybercrimes such as identity theft, fraudulent account openings, and financial scams increases. To address the rising threat of deepfakes, this research explores detecting deepfake face images by combining transfer learning with an ensemble technique. Four pre-trained models have been employed for the transfer learning task. Finally top three performing models were combined for the ensemble. The ensemble model has been evaluated against a benchmark dataset, namely 140K Real and Fake Faces. The ensemble model significantly surpassed the individual models, achieving an accuracy of 81.25%. This research demonstrates the potential of integrating multiple pre-trained models to improve deepfake image detection, laying a strong foundation for future advancements.

Enhancing Topic Modeling Through Embedding Learning Strategies

Pallabi Biswas — 2025-06-20

In the field of Natural Language Processing (NLP), topic modeling is crucial for uncovering patterns in textual data. Recent advances have combined traditional topic modeling with word embeddings, introducing the Embedded Topic Model (ETM). This thesis explores embedding learning strategies within topic modeling to improve the ETM and related models. It delves into more efficient variational inference, advanced word embedding techniques, and strategies for better topic interpretability. Practical implications in document classification, content recommendation, and summarization are evaluated. Scalability challenges for handling large textual corpora are also addressed. The integration of textual data with other modalities is pioneered. This work aims to enhance topic modeling using embedding learning strategies, bridging the gap between theory and practice in NLP.