The Covid-19 outbreak has adversely influenced university students across the world both physically and psychologically. The psychological struggle faced by students, is effected by various factors, including disruptions to daily routines and academic activities, increased reliance on smartphones and the internet, limited social interaction, and confinement to their homes. These impediments reflect a broader issue of imbalance in cognitive health status among them during the pandemic. In Bangladesh, despite having the necessary population to study, understanding the impact of Covid-19 on the mental health status of university students has been limited. Hence, it is imperative to diagnose mental health issues and deal with the underlying reasons in order to enhance students’ psychological well-being, which leads to academic achievement. Nowadays, Artificial Intelligence (AI) based prediction models have the potential to play a crucial role in predicting mental state early. The purpose of the study is to explore the following effects of the pandemic on the mental health of Bangladeshi university students using Machine Learning (ML) and Deep Learning (DL) techniques. A reliable AI prediction system requires real-world data, that was collected by a survey through a Google form (online questionnaires) among 400 students of 16 universities, and the respondents were 253. In this paper, after data preprocessing, ten widely known ML and four DL models were developed to automatically and accurately predict mental well-being during or after the Covid-19 circumstance. According to our findings, the Random Forest (RF) algorithm and Siamese Neural Networks (SNNs) outperformed other models in terms of accuracy (86% and 75%). Additionally, Chi-Square test was conducted, which revealed the five most common and significant predictors (“Stable family income”, “Disruption of daily life”, “Own income”, “Sleep status”, and “Fear of getting infected with Covid-19”) of psychological health conditions. Overall, this work could assist university administrations, government agencies, and health specialists in taking appropriate measures to understand and maintain students’ mental health. This research also suggests proper monitoring, government support, and social awareness during and after the worldwide epidemic to keep an excellent mental health state of university students.

The classification of Bangla characters and digits is a fundamental component of Natural Language Processing (NLP) and computer vision applications. However, despite advancements in handwritten character recognition (HCR) for other languages, recognizing Bangla script remains a formidable challenge, primarily due to its extensive character set, intricate compound forms, and significant stylistic variations. While existing datasets have significantly advanced the field of Bangla HCR, they frequently overlook the complexity and variability of primary-level learners' handwriting. This paper introduces the first extensive Bangla Primary Script 2025 (BPS2025) dataset, a novel, balanced and comprehensive, demographically oriented collection of isolated characters and numerals specially focused young primary school students. The dataset was selectively curated from 500 students, aged 7 to 12 and in grades 2 to 5, across four districts in Bangladesh. It comprises 24,420 raw images across 60 balanced classes, including 50 basic characters and 10 digits, which followed five stage pre-processing pipelines to process the final dataset. The dataset addresses a significant gap in existing benchmarks, which is expected to support future research and real-world educational applications by facilitating the development of robust and accurate recognition models for Bangla script.

Background: Cardiovascular disease (CVD) remains a leading cause of global mortality, with many risk factors being potentially modifiable, highlighting the importance of early risk assessment. Machine learning (ML) can support accurate CVD risk prediction and provide explanations for individual predictions. However, the practical use of explainable AI (XAI) remains limited by unstable feature rankings across training samples, limited distinction between modifiable and non-modifiable factors, and insufficient verification of whether recommended changes reduce model-estimated risk. Objective: To bridge these gaps, this study introduces pCARE-CV, a unified XAI framework that integrates predictive modeling, bootstrap-based SHapley Additive exPlanations (SHAP) stability analysis, a SHAP-guided recommendation module (SGRM), and multi-method counterfactual (CF) analysis within a single, deployable, and clinically interpretable cardiovascular risk assessment system. Thus, our work extends the methodological foundation of interpretable ML for applications in biomedical. Methods: Seven ML classifiers were evaluated on two synthetic cohorts, EarlyMed (n = 70,000) and HDD-25 (n = 3,069), and one real-world cohort, UCI-HDD (n = 920), using multiple predictive performance metrics with bootstrap confidence intervals. SHAP ranking stability was assessed through 200 bootstrap refits using Spearman correlation, Kendall’s tau, and top-k overlap. Multi-Objective Counterfactuals (MOC), Nearest Instance Counterfactual Explanations (NICE), and What-If CF methods were compared on EarlyMed and UCI-HDD using validity, sparsity, proximity, plausibility, and predicted probability reduction. The proposed framework is demonstrated through a Streamlit-based web application that enables real-time translation and provides an applicability and feasibility recommendation. Results: CatBoost achieved accuracies of 99.36%, 63.03%, and 85.87% on EarlyMed, HDD-25, and UCI-HDD, respectively. Mean bootstrap Spearman correlations of SHAP rankings were 0.886, 0.732, and 0.837, indicating differences in attribution stability across datasets. On EarlyMed, MOC achieved the highest validity (1.000) and mean probability reduction (0.7075), while What-If achieved the highest plausibility (1.3255). NICE generally required fewer feature modifications but showed slightly lower validity. On UCI-HDD, the relative performance of the CF methods varied across evaluation criteria, underscoring the potential applicability of CFs in clinical settings. Conclusion: pCARE-CV integrates prediction, explanation stability, actionable recommendations, and CF analysis within a single deployable framework. The results highlight the importance of evaluating predictive performance and explanation stability across datasets, as well as CF explanations in medical domains using multiple complementary criteria. Together, these evaluations can support more nuanced and effective cardiovascular patient care strategies.

Accurate depth estimation is a fundamental requirement in underwater vision, but it is severely hindered by haze, low contrast, and colour distortions that degrade image quality and obscure structural details. To address these challenges, we propose AquaDepth, a novel three-stage hybrid framework for depth prediction that incorporates image enhancement, restoration, and depth estimation. In the first stage, underwater images are enhanced using de-hazing, contrast-limited adaptive histogram equalization (CLAHE), and edge refinement to improve visibility and structural clarity. In the second stage, a generative restoration process, including stable diffusion, is applied to recover subtle object boundaries and fine textures often overlooked or vague in conventional approaches. Finally, depth prediction is performed using a transformer-based vision model, ensuring that both global clarity and fine structural details contribute to reliable perception. Evaluation on the UIEB and LSUI datasets demonstrates the effectiveness of the proposed framework. For the enhanced and restored RGB images, the method achieves an NIQE score of 3.02 and a UIQM score of 62.94 on UIEB, and an NIQE score of 2.98 and a UIQM score of 59.78 on LSUI, outperforming existing state-of-the-art enhancement approaches. Beyond RGB quality improvement, the perceptual quality of the generated depth maps is also quantitatively evaluated. After the complete enhancement-restoration pipeline, the resulting depth maps obtain an NIQE score of 13.88 and a UIQM score of 71.97 on UIEB, and an NIQE score of 13.65 and a UIQM score of 37.89 on LSUI. Both quantitative and qualitative analyses indicate that the proposed approach significantly improves image visibility and depth representation, producing depth maps with clearer structural boundaries, better depth separation, and higher perceptual consistency, thereby enabling more accurate and visually reliable depth perception.

The transportation sector is the primary consumer of vehicle fuel worldwide and is thus a major contributor to climate change via carbon dioxide (CO₂) emissions. In addition to severe environmental impacts, such as global warming, droughts, floods, and rising sea levels, these emissions have a negative effect on public health by increasing the prevalence of respiratory disease. Achieving environmental sustainability through regulatory oversight requires a strong understanding of vehicular fuel consumption and CO₂ emissions. However, accurate modeling of these remains challenging due to the complex non-linear relationships between various vehicular characteristics and the lack of interpretability of many predictive models. Traditional linear models often fail to capture high-dimensional data complexities, while black-box methods provide few actionable insights for policymaking. To address these gaps, we developed a robust and data-driven two-stage machine-learning (ML) framework designed to enhance model performance and reliability. First, we implemented standard data preprocessing, enhanced feature engineering, and hyperparameter tuning for 14 cutting-edge ML algorithms and three advanced modeling techniques to explore their predictive performance. Second, we introduced three interpretable explainable AI (XAI) approaches. These were evaluated on a publicly available Kaggle static dataset of 550 vehicles, dominated by gasoline-powered vehicles, with only two diesels and two electric vehicles. The tuned CatBoost model demonstrated strong predictive performance, achieving an impressive R² of 0.9260, a root mean square error (RMSE) of 1.1759, and a mean absolute error (MAE) of 0.8147. In parallel, we deterministically estimated CO₂ emissions from fuel consumption, which provide direct estimates of tailpipe emissions. To ensure transparency and model interpretability, we employed Shapley additive explanations, local interpretable model-agnostic explanations, and permutation importance to identify the key factors contributing to the model predictions. Across the explainability analyses, cylinder count, front-wheel drive (drive_fwd), and the displacement–year interaction were the primary contributors to the predicted combined miles per gallon; in other words, they strongly affected fuel consumption. Collectively, these findings demonstrate the ability of the proposed model to capture complex feature relationships; thus, it offers a valuable tool for researchers and policymakers in sustainability planning and emission control. Future research should focus on real-time driving or dynamic measurements data and enhancing practical applications to further reduce emissions and promote environmental sustainability.