Journal of Computing & Biomedical Informatics

Clustering Student Performance: A Data-Driven Approach to Monitor Academic Success

Rohma Qadir, Areej Fatemah Meghji, Naeem Ahmed Mahoto — Mon, 01 Dec 2025 00:00:00 +0000

The goal of Educational Data Mining (EDM) is the exploration of hidden patterns and insights in educational data. Making use of the EDM approach of clustering, this research explores the analysis of variation in students performance across the course of an academic degree. We perform experiments on the data of 210 students belonging to the Department of Software Engineering in an attempt to discover patterns between three class of learners’ – high performers, intermediate performers, and low performers. These patterns are not only analyzed across different learner classes but also across different genders. The research also makes use of heatmap analysis to highligh subject-wise performance and to better understand the subjects that students struggle in. The findings of the reseach highlight the subjects that students have difficulties in and show that although students in most instances performed well in theoretical courses, several students had difficulty in practical courses. A comparison between two batches revealed that Batch-02 had generally improved performance which was particularly evident in the sixth semester of the degree program. These findings provide an alternative understanding of the intricate interaction between academic performance and student behaviors, which can be invaluable in guiding educators and policymakers to devise interventions that could help students achieve better results and ultimately reshape the learning paradigm.

A Practical Analysis of the Fundamentals of Sensory Immersion through Heavy Plasticity of Stage Appearance

Hyuntai Kim — Mon, 01 Dec 2025 00:00:00 +0000

The given paper analyses the basics of the processes that make it possible to achieve the sensory immersion based on the combination of the newest technologies in the sphere of audio and the active design of the visual stage. Our approach to testing the quantitative approach of immersive experiences is the interaction of spatial audio systems, adaptable lighting structures, and physical changes to the stage. Through controlled experiments of 240 participants in 12 different performance settings we formulate mathematical models of the correlation between technological variables with measurable immersion metrics. This is shown to be a key indication of increasing sensory involvement by our result that the so-called heavy plasticity of stage appearance that is marked by the ability to dynamically and malleably modify both visual and spatial items is particularly important when it is accompanied by multi-dimensional sound fields. The paper illustrates that, the immersion intensity (I) has a power-law relationship with technological integration density (0 tech ) and sensory coherence (Cs ), which are represented by the equation I = k 0 tech a Cs_b where a = 1.37 and b = 2.14. The results suggest that maximum immersion is possible when the use of sensory stimuli is 6.8 per second and the spatial coverage of audio is more than 85% of the performance area. The present study offers empirical bases of development of next-generation immersive experiences in the entertainment, virtual reality, and therapeutic applications.

Comparative Analysis of LSTM-Based Variant Models for Detecting Attacks in IoT Networks

Mosleh Abualhaj — Mon, 01 Dec 2025 00:00:00 +0000

Internet of Things (IoT) networks have established unparalleled connection possibilities and convenient features and have set new challenges associated with dubious security barriers and potential attacks. This study evaluates attack detection performance of LSTM-Based Variant models namely LSTM, DeepBiLSTM and BLSTM recurrent neural networks by bringing up experiment analysis of the IoT networks. We evaluate the three models using benchmark IoT dataset in terms of detection accuracy, precision, recall and F1 measure. The experiment results indicate that the three LSTM-based models, LSTM, BiLSTM, and DeepBiLSTM, demonstrate a high level of performance, regardless of the batch size (32, 64, 128, 256, and 512). DeepBiLSTM has a little better overall performance, which validates its soundness and suitability towards scale in detecting attacks in large IoT networks.

Systematic Review of AI-Based Approaches for Anthropometric and Fashion Landmark Detection in Body Measurement Estimation

Aqsa Jameel, Tanzeela Kousar — Mon, 01 Dec 2025 00:00:00 +0000

This systematic review gives a detailed discussion of anthropometric landmark abstraction and dimension measurement techniques with a focus on their use in the Fashion and Apparel (F&A) industry. It starts with an overview of the leading object detection models and their application in the detection of garments and human body features. The review makes a clear distinction between fashion landmark detection, which aims at detecting key points on clothing, and anthropometric landmark detection, which isolates anatomical landmarks on the human body to obtain measurement estimates. Different measurement extraction techniques are addressed, which include 2D silhouette analysis, 3D body scanning, and mesh-based modeling to acquire standardized anthropometric parameters, which include lengths, breadths, depths, and circumferences. The originality of this review is that it is the first analytical framework that combines the two domains of anthropometric and fashion landmark detection that have been historically examined separately. The review fills the gap between the human body measurement estimation and clothing landmark abstraction by providing a cross-domain synthesis, which provides a unified view of algorithms, datasets, and evaluation metrics. Moreover, it compares new methods including classical machine learning methods and modern deep learning and ensemble methods, demonstrating the performance increase depending on the accuracy metrics like Mean Absolute Error (MAE) and Normalized Error (NE). The review identifies a clear research direction that is moving away the conventional computer vision pipelines to the data-driven deep learning solutions. In general, the review provides new knowledge that can be used to develop garment fit prediction, virtual try-on technologies, and intelligent apparel recommendation systems by incorporating anthropometric and fashion-based landmark detection strategies.

Design and Implementation of a Virtual Reality Experience System Incorporating Popular Music and Dance

Huaping Kong, Hyuntai Kim — Mon, 01 Dec 2025 00:00:00 +0000

In this paper, we introduce a design of immersive virtual reality system that combines pop music and dance to attract the user and bring them the entertainment experiences. Using motion capture, real time audio analysis, and 3D visualization, the proposed approach generates an interactive scenario in which users learn, perform, and experience dance choreography to pop music. The system architecture is based on the Unity3D game engine and the Oculus Quest 2 VR headset, combined with in-house motion tracking algorithms to offer a dynamic and fun experience. We investigated learnability, immersion, and learning effectiveness through two user studies involving three age groups with 45 participants in total. The VR dance app scored 78.4 on average in the SUS. The74% of respondents reported to have improved significantly in dancing, and 83% to feel they were performing better. The system latency was limited to less than 20ms to maintain synchronization of user’s movement and feedback in the virtual world. This work advances the development of VR-based entertainment and education, and establishes that music and dance may be leveraged as sources of fun and learning in virtual worlds.

Multivariate Air Pollution Anomaly Detection via LSTM Autoencoders on Beijing Multi-Site Sensor Data

Mon, 01 Dec 2025 00:00:00 +0000

The current paper investigates the application of a Long Short-Term Memory Autoencoder (LSTM-AE) to identifying anomalies in the multi-variate time-series of air-pollution measurements. The algorithm was used to the sensor data of PM2.5, CO, O3, NO2, TEMP, and WSPM of the Beijing Multi-Site Air-Quality Data Set described on Kaggle. The test set included fifty synthetic anomalies that were used to assess performance. The anomalies were identified using the reconstruction error calculated using Mean Squared Error (MSE) as a dynamic threshold value of 0.002957. The presented model produced the Precision, Recall, F1-Score, and ROC-AUC of 77.00%, 100.00%, 87.01%, and 99.86%, respectively, proving its effectiveness in detecting minor and drastic changes in the pattern of air quality.

From Sampling to Auto-Tune: Intergenerational Divergence in R&B Vocal Production and Audience Acceptance

Qian Xie, Hyuntai Kim — Mon, 01 Dec 2025 00:00:00 +0000

This article examines the relationship among technological progress in R&B music production (e.g., sampling technology, Auto-Tune usage) and intergenerational audience acceptance patterns. Applying a mixed-methods design that combines quantitative audience surveys (N=847) across four generational cohorts (Baby Boomers, Generation X, Millennials, Generation Z) with acoustic analysis of 240 R&B songs spanning 1980-2023, we explore how vocal rhythm design evolved in conjunction with these technologies, and how different age groups react to such changes. The findings reveal notable age-based differences in acceptance, with more favorable audience (Gen Z: 78.4% acceptance) opinion towards Auto-Tune usage on music recordings when compared to older generations (Baby Boomer: 31.2% acceptance). Sampling technology has more intergenerational playability (64.7% overall playability) than Auto-Tune (52.3% overall playability). The research demonstrates that the patterns of acceptance are closely influenced by perceptions of technological authenticity, the length of exposure, as well as by cultural context. This extends previous research on how production technologies, when applied to genres, create new subgenres and specific audience segments in the context of contemporary R&B music.

Intelligent Firewall for Attack Detection: Integrating Dragonfly and Bat Algorithms with Machine Learning

Ali Al-Allawee, Sultan Aldossary, Radhwan M. Abdullah, Lway Faisal Abdulrazak — Mon, 01 Dec 2025 00:00:00 +0000

The increasing sophistication of cyber threats necessitates the development of advanced attack detection methods capable of handling high-dimensional network traffic data efficiently. This paper introduces an AI-driven firewall model that leverages the Dragonfly Algorithm (DA) and Bat Algorithm (BA) for optimal feature selection, enhancing attack detection accuracy. The proposed approach utilizes the UNSW-NB15 dataset and employs a union-based feature selection strategy, combining the best-selected features from DA and BA to maximize classification performance. Three classifiers— utilize Decision Tree (DT), Support Vector Machine (SVM), and Logistic Regression (LR)—are implemented for attack detection. Experimental results demonstrate that DT achieved 100% accuracy, SVM achieved 99.99% accuracy, while LR achieved 99.94%, confirming the effectiveness of the proposed model. The AI-embedded firewall significantly reduces false positives and enhances detection robustness.

AI Safety and Trustworthiness: A Survey

Ahmad Raza, Irshad Ahmed Sumra, Abdul Sattar — Mon, 01 Dec 2025 00:00:00 +0000

Artificial intelligence (AI) is now incorporated into many important areas of concern, its choices and actions may directly affect people's lives in a variety of fields, including healthcare, economics, education, and even government. The rapid adoption of AI raises questions about safety, dependability, and credibility despite some of its incredible skills in automation, pattern recognition, and problem-solving. The topic of AI safety has become a global concern because to unintended outcomes, including bias, adversarial resistance, a lack of transparency in decision-making, and irrelevance to human values. The safety and reliability of AI will be discussed in this article from a technical, ethical, and governance standpoint. It investigates how to create AI systems that consumers and other stakeholders can trust by utilizing robustness, security, transparency, fairness, and accountability. By examining the present frameworks, research, and legislation, the study identifies the issue of striking a balance between innovation and safety. It also suggests the path of future study, such as the accountability of AI implementation, human-centered development, and interdisciplinary collaboration. The need to create safe and reliable AI is discussed in the paper as both a technological and, more importantly, a socio-ethical issue that calls for cooperation from academics, business, government, and civil society.

Harnessing Deep Learning and Language Models for Protein Function Prediction: A CAFA5-Based Study

Ali Haider, Jamal Shah, Musadaq Mansoor, Omar Bin Samin — Mon, 01 Dec 2025 00:00:00 +0000

Recent advancements in deep learning have brought remarkable progress in the area of predicting the protein functions from its amino acid sequences. These sequences play a crucial role in accelerating drug evaluation and uncovering how cells work. This research investigated several deep models for predicting protein functions, which include Bi-LSTM coupled with an attention mechanism, Gated Recurrent Unit, Long Short-Term Memory, Deep Neural Networks, and Bidirectional LSTM. This research used the CAFA5 dataset along with the T5 embedding, which is created from this dataset, to test these DL models for the multi-label protein functions prediction task. The researchers used state-of-the-art matrices to measure the performance of these models, which includes ROC-AUC, Hamming loss AUC, and binary accuracy. The analysis demonstrates the Bi-LSTM paired with attention mechanism and DNN models outperformed the baseline traditional RNN models in both minimizing loss and accuracy. With an outstanding ROC-AUC score of 0.9239 and consistent prediction reliability, the Bi-LSTM plus Attention model performed well. This research showed that combining DL models with integrated attention layers produces more scalable and accurate results for predicting protein functions. Showing their usefulness in practical bioinformatics tasks.

Causal and Explainable Machine Learning Framework for Heart Disease Prediction using XGBoost and SHAP

Mon, 01 Dec 2025 00:00:00 +0000

Cardiovascular disease constitutes one of the major leading causes of deaths in the globe. Early diagnosis is needed to enhance patient outcomes.Although machine learning models, including XGBoost, are highly accurate in predicting heart disease, they are black-box and therefore cannot be interpreted clinically.To overcome this shortcoming, we devised a new model that integrates: XGBoost and SHAP(SHapley Additive exPlanations), which yields the impact of each feature on the prediction. Using the PC (Peter-Clark) algorithm, we determined the causal relationship between features and the outcomes of heart diseases and differentiated causation with correlation.To have the system useful in real-life healthcare, we created a simple interface allowing doctors to input patient information, view predictions, and read about explanations in various levels of detail.Our model was tested on the UCI Heart Disease dataset and achieved 91% accuracy, 0.90 F1-score, and 0.95 AUC-ROC better than other common models (Logistic Regression, Decision Tree, and Random Forest). Our tool will assist doctors in making better judgments regarding the risk of heart disease by integrating good predictability, explicit explanations, and user-friendly design.