http://dspace.uiu.ac.bd/handle/52243/29 Sun, 12 Apr 2026 21:08:45 GMT 2026-04-12T21:08:45Z http://dspace.uiu.ac.bd/handle/52243/3408 A Secure Spectrum Aware MAC layer protocol for Cognitive Radio Wireless Sensor Network. Hossain, Md Shamiun Persistent Secure Spectrum Aware MAC layer, scalability and load balancing are important requirements for numerous ad-hoc sensor network. Secure Clustering sensor nodes is an effective technique for achieving these goals. In this work, we present a secure, spectrum-aware cross-layer MAC protocol designed for Cognitive Radio Ad Hoc Networks (CRAHN), where cluster formation is defined maximum edge biclique problem, enhance the security of the cluster formation process and ensure both a stable number of common channels and robustness to varying spectrum availability. The clustering process concludes in O(1) of iterations, independent of the network’s structure or scale. By carefully choosing the secondary clustering parameter, the process can significantly reduce the hassle of re-clustering. This thoughtful selection ensures that the workload is evenly spread across the cluster heads, preventing any one cluster from becoming overloaded, and keeping the network running smoothly and efficiently. Dynamic approach maintains a stable network structure despite node mobility. These protocols focus on secure and efficient spectrum sharing among nodes to enhance network performance. It also achieves fairly uniform cluster head distribution across the network. The cluster-based architecture is designed to be highly flexible, allowing it to build and adjust itself as needed. This dynamic nature is supported by two key operations: Node-Move-In and Node-Move-Out. These operations enable the network to easily integrate new nodes and remove existing ones, ensuring that the system can adapt to changes and maintain optimal performance without manual intervention. A pseudocode analysis focused on enhancing the security of the cluster formation process is also applied in two topology management operations. Overall, the time complexity for the Node-Move-In and Node-Move-Out algorithms is O(n), where n represents the number of members in the cluster. This means that the algorithms handle a number of operations proportional to the number of nodes involved, making them efficient and scalable as the network grows. Tue, 31 Mar 2026 00:00:00 GMT http://dspace.uiu.ac.bd/handle/52243/3408 2026-03-31T00:00:00Z http://dspace.uiu.ac.bd/handle/52243/3393 A Hybrid Approach to Bangla Regional Text Classification Using BERT Ensemble and Region-Specific Lexical Oversampling Sultana, Babe Regional text analysis reflects the lived realities of diverse communities by capturing the linguistic richness and diversity present in various dialects. It bridges the gap between everyday regional usage and standardized language forms, thereby enhancing the inclusivity of language technologies. In this paper, we focus on five regional dialects in Bangladesh, namely Chittagong, Sylhet, Noakhali, Barishal, and Rangpur, using a dataset of 4,218 text samples. The dataset is validated by five regional experts and categorized into three tiers based on an assigned agreement criterion. Tier 1 represents a strictly filtered, high-confidence subset and is used primarily for evaluation. A set of region-specific special words, which belong exclusively to their respective regions and are validated by domain experts, is introduced. These words are used in a linguistically informed oversampling technique to balance the dataset in both experiments. In the first experiment, we demonstrate the effectiveness of the tiered dataset structure, where Tier 2 and Tier 3 (mediumand low-confidence subsets) are used for training, and Tier 1 (high-quality subset) is used for testing. In this setting, BanglaBERT achieves the best individual performance with 67.45% accuracy and a weighted F1-score of 67.62%. In the second experiment, we focus exclusively on the Tier 1 dataset, applying a wide range of machine learning and deep learning models to assess their effectiveness. The key contribution is a heterogeneous deep ensemble technique that combines three BERT models, BanglaBERT, BUETBERT, and DistilBERT, achieving an accuracy of 85.17% and a weighted F1-score of 84.84% on the Tier 1 dataset. CSE Mon, 12 Jan 2026 00:00:00 GMT http://dspace.uiu.ac.bd/handle/52243/3393 2026-01-12T00:00:00Z http://dspace.uiu.ac.bd/handle/52243/3392 An Explainable Ensemble Convolutional Neural Network for Early Lung Cancer Prediction with Web Application Noor, Kazi Rifah Lung Cancer is one of the deadliest forms of cancer significantly contributing to the rising mortality rates globally. The high mortality rate associated with lung cancer can largely be attributed to its late detection, as symptoms often do not appear until the disease has reached advanced stages. Early detection of anomalies in medical imaging, particularly at the initial stages, is crucial for advancing both quantitative image analysis and patient care. In this context, our research introduces a fully automated web application to predict lung cancer early on CT scan images. This application leverages the power of a weighted average ensemble-based deep learning framework that combines multiple neural network architectures to enhance the reliability of automated classification. The work highlights the need for interpretability in clinical decision support systems, going beyond classification. Our proposed approach is organized into three independent phases. First, we performed image augmentation as part of the preprocessing stage analyzing the IQ-OTH lung cancer dataset which implies that the model is trained on diverse and enriched input data. The system incorporates ResNet50, VGG16, and a custom CNN, all of which provide complementary feature-learning capabilities that improve overall predictive dependability. The ensemble model clearly outperformed the individual networks, achieving an accuracy of 92.28% on the IQ-OTH lung cancer dataset. The visual indications and regions that most influence our model’s decisions are highlighted using Explainable Artificial Intelligence techniques, particularly LIME and SHAP. By making the model’s inner workings more understandable, these explanations hope to boost medical practitioners’ confidence. Finally, the ensemble model was integrated into a user-friendly web application allowing users to upload CT scan images, which are then analyzed to classify the images into normal, benign, or malignant categories. Furthermore, our web application gives confidence levels for each prediction increasing the credibility of its results. Our system provides medical practitioners with a smooth and effective tool by integrating the precision of an ensemble model with the flexibility of a web application. CSE UIU Mon, 12 Jan 2026 00:00:00 GMT http://dspace.uiu.ac.bd/handle/52243/3392 2026-01-12T00:00:00Z http://dspace.uiu.ac.bd/handle/52243/3391 Recognition of Bangla and English Words in Bangla Texts Using a Modified BERT-base-NER Model Hossain, Md. Parvez A combination of Bangla and English words is commonly used, particularly on social media. This tendency greatly hampers the next generation’s ability to learn Bangla. This study suggests an approach for identifying words in Bangla texts that are both English and Bangla. This study also translates the identified English terms into standard Bangla words. The Transformer architecture, which uses an attention mechanism to identify the connections between words and their contexts inside a text, is the foundation of bidirectional encoder representations from transformers (BERT). In this study, we use the training input dataset to modify the BERT-base-NER model. For the name entity recognition (NER) task, the proposed BERT-base-NER model in this study achieves state-of-the-art performance. For both the training and testing scenarios, we employ a holdout cross-validation procedure. We used 80% of the entire data for training and 20% for testing. We use the Google Translate API (application programming interface) to translate the identified English words into standard Bangla words. In order to assess the modified BERT-base-NER model, we applied the input dataset to the current machine learning (ML) and deep learning (DL) techniques. Support vector machines (SVM) and Naive Bayes (NB) are two components of the machine learning approach. Conversely, the DL method uses bidirectional LSTM (BiLSTM), long short-term memory (LSTM), and convolutional neural network (CNN). The improved BERT-base-NER model is highly accurate and efficient at identifying Bangla and English words, according to simulation data. With an accuracy of 95%, the proposed BERT-base-NER model achieves the best result among the current methods. For Bangla–English code-mixed text, this study presents a reliable BERT-based word-level language identification system that successfully resolves Banglish ambiguity and allows downstream Bangla language processing applications such as standard Bangla conversion, machine translation, and information extraction. CSE UIU Mon, 12 Jan 2026 00:00:00 GMT http://dspace.uiu.ac.bd/handle/52243/3391 2026-01-12T00:00:00Z