Jahangirnagar University Journal of Electronics and Computer Science

Expectation Maximization Algorithm: A Review

Md. Masum Bhuiyan — 2026-08-25

The Expectation-Maximization (EM) model is used to evaluate the maximum likelihood of parameters of a statistical model using an iterative method. The EM algorithm is applicable when the given data set is incomplete, i.e., data has two parts: one is the observed variable (known), the other is the latent variable. Latent (hidden) variables are not directly observed but inferred from observed variables using some mathematical model. Before using the EM algorithm, we need to understand observed variables, latent variables, likelihood, and Jensen’s inequalities. In this review work, the basic theory of all the mathematical models used in the implementation of the EM algorithm, operational steps of the algorithm with explanation, and numerical examples are shown explicitly to get a clear concept of it.

Detection Of Corn Leaf Diseases Using Convolutional Neural Network And Transfer Learning

Md. Abdul Mukit — 2026-08-25

This study seeks to develop a neural network model for the identification of corn leaf diseases. Corn leaf diseases substantially affect agricultural yield and food security. Three different types of corn leaf diseases are caused by pathogens, and fungi drastically hamper the production. Without proper action, the entire crop field can be infected, resulting in immense loss in the economy. By consuming affected crops, many diseases can get into the human body. This also increases the probability of getting cancer, which is fatal for humans. In order to address this issue, this study suggests a deep learning-based multiclass classification approach that uses transfer learning (TL) and convolutional neural networks (CNNs) to identify different maize leaf diseases. A dataset of infected and healthy corn leaves is used to train the CNN model. This dataset has been gathered from Kaggle and other sources. Two supervised learning methods are used, and the performance of the models is compared. A pre-trained DenseNet121 transfer learning model is used to enhance the performance of our research. To identify leaves from four different classes, we employed softmax activation in our neural network’s output layer. We got 93.4% accuracy in the convolutional neural network model, and 98.21% accuracy in DenseNet121 in the evaluation phase. This model can efficiently be applied in our agricultural production. Compared to the old native ways, this process requires less time and provides a dynamic way in the detection of corn or maize diseases.

Segmentation Strategies for Long-Code Authorship Detection: Balancing Context Coverage and Computational Efficiency

Mehedi Hasan — 2026-08-25

Detecting the authorship of source code becomes substantially more difficult when programs exceed the fixed context window of transformer-based models, as important stylistic and structural cues may fall outside the observed input. Although long programs are common in real-world software repositories, there is limited systematic understanding of how practical segmentation strategies influence authorship detection under such constraints. We investigate this problem by comparing three widely applicable approaches that require no modification to pretrained models: truncation-based classification, multi-view cropping with aggregation, and sliding-window classification with logit pooling. These methods reflect different trade-offs between computational cost and context coverage. All approaches are evaluated under a unified experimental setup on a lengthstratified dataset that emphasizes long code samples. Multi-view classification substantially improves over truncation, achieving 0.896 accuracy and 0.860 macro-F1, compared to 0.862 accuracy and 0.817 macro-F1 for truncation (+3.43 and +4.31 points, respectively). Sliding-window classification attains the highest overall performance (0.897 accuracy, 0.862 macro-F1), but multiview reaches near-parity (within 0.10 accuracy and 0.23 macro- F1) while requiring far fewer forward passes. Improvements are most pronounced for hybrid and adversarial code, where discriminative evidence is dispersed across distant regions of the program. These results indicate that context coverage is a primary factor in effective code authorship detection and provide practical guidance for selecting efficient long-context handling strategies when full-program processing is infeasible.

Hand Landmark Detection Using YOLO26n for Gesture-Based Smart Wheelchair Navigation

Rakib Ahammed Diptho — 2026-08-25

Assistive mobility technologies enhance the independence of individuals with physical disabilities. Within its scope, the gesture-based control offers a particularly intuitive solution by enabling contactless interaction through computer vision. With this concept, we propose a smart wheelchair based on the YOLO26n-pose deep learning model for accurate hand keypoint detection. The model is trained on the Hand-Keypoints dataset, which contains 26,768 annotated images with 21 hand landmarks, enabling robust hand landmark detection that performs better than many existing approaches. Using the detected keypoints, a distance-based gesture recognition method analyzes finger movements and generates wheelchair navigation commands. Experimental results demonstrate a strong detection performance, achieving a mean average precision of 0.992 with an inference time of 2.6 ms, making the system suitable for real-time wheelchair operation. The fast and precise processing capability of the proposed approach outperforms the contemporary methods in terms of speed, efficiency, and reliability.

An Optimized Hybrid Approach for Code Smell Detection Using Machine Learning and Attention-Based FNN

Taniz Fatema — 2026-08-25

Code smells are structural flaws in the source code that have a detrimental effect on readability, maintainability, and dependability. To adress this challenges we proposed a hybrid model using ML classifiers with attention-based Feedforward Neural Network. In this work, we applied five ML classifiers (SVM, KNN, LR, DT, RF), two feature selection techniques are RFECV and Information gain on different datasets namely DC, FE, GC, LM. To enhance the performance and reduce the overfitting-underfitting issues we applied two optimization techniques are Bayesian Optimization and Grid Search while applying Information Gain feature selection technique. For deeply analyzing trained the model with attention-based Feed Forward Neural Network. Experimnetal outcome shown that our proposed model obtained the highest accuracy on Long method dataset, 99.69% by using SVM, RF classifiers and Information gain with Bayesian optimization technique. Also achieved the accuracy, 99.68% for RF, DT, LR classifiers with RFECV feature selection technique on same dataset.

Style-Aware Bangla News Headline Generation Using Large Language Models: A Comparative Zero-Shot Evaluation

Lubna Yasmin Pinky — 2026-08-25

News headline generation is an essential task in natural language processing (NLP), as it plays an important role in conveying the main message of a news article in a concise form. Although this task has been widely studied in English and other high-resource languages, Bangla headline generation remains less explored, especially when the headline needs to follow a specific writing style. In this study, four Large Language Models (LLMs), namely GPT-OSS-120B, LLaMA-3.3-70B, Qwen3-32B, and Qwen2.5-14B-Local, are evaluated for style-aware Bangla news headline generation in a zero-shot setting. The models were not fine-tuned for this task; instead, they were guided only through style-specific prompts. For the experiment, 75 Bangla news articles were used, and headlines were generated in six styles: Political, Informative, Dramatic, Short, Analytical, and Event-Based. A total of 1,800 generation attempts were performed. The generated headlines were evaluated using 15 automatic metrics, including BLEU-1/2/4, METEOR, ROUGE-1/2/L, WER, CER, Edit Distance, TF-IDF Cosine Similarity, Jaccard Similarity, and BERTScore Precision, Recall, and F1. The results show that LLaMA-3.3-70B achieves the best overall performance, with a BLEU-1 score of 0.1454, a METEOR score of 0.1333, and a BERTScore-F1 score of 0.2064. Qwen3-32B obtains the highest BERTScore Recall score of 0.2144, indicating better coverage of the reference headline content. The results also indicate that the Informative style produces more accurate and consistent headlines than the other styles. Overall, this study demonstrates the potential of zero-shot LLMs for style-aware Bangla headline generation, while also highlighting the limitation of using a single reference headline for evaluating creative generated outputs.