Accepted in Scientific Reports. Across 29 cities, a zero-shot time-series foundation model matches a tuned specialist on PM2.5 and temperature once covariates are restricted to what is knowable at forecast time, and uses about 10× less measured energy under frequent retraining.
Publications
8 papers · See Google Scholar for up-to-date metrics
Comparative analysis of machine learning techniques for cardiovascular disease prediction
Md. Mahfuz Uddin , Md. Binyamin , , Md. Rezaul Karim
Benchmarking eight ML algorithms on 61,711 patient records, LightGBM comes out on top for cardiovascular disease prediction (AUC 0.800), with systolic blood pressure, age, and cholesterol as the strongest predictors.
Green-NAS: A Global-Scale Multi-Objective Neural Architecture Search for Robust and Efficient Edge-Native Weather Forecasting
, Soyda Humyra Yesmin , Saiful Islam , Md. Palash Bin Faruque , Md. A. Salam , Md. Mahfuz Uddin , Samiul Islam , Tofayel Ahmed , Md. Binyamin , Md. Rezaul Karim
153k parameters, 239× fewer than GraphCast, with RMSE within 1.4% of our manually tuned baseline. Multi-objective NAS finds deployable models, and transfer learning adds ~5% accuracy when historical data is scarce.
Predicting Early Childhood Development in Bangladesh Using Hybrid Machine Learning Model and SHAP Explainability
Saiful Islam , Md. Palash Bin Faruque , Tanjina Khan , Most. Shabrina Afroz , Tofayel Ahmed , , Md. Kamruzzaman , Md. Mostafizur Rahman , Md. Abdul Khalek
Hybrid model (SVM+RF+XGBoost) predicts early childhood development in Bangladesh with 0.77 accuracy (cognitive) and 0.71 (social-emotional). SHAP reveals domain-specific drivers: books and education for cognition, caregiving and discipline for social-emotional outcomes, with urban-rural subgroup divergences informing targeted SDG 4.2 interventions.
Temporal Validation Changes the Apparent Public-Health Utility of Under-Five Mortality Prediction in Bangladesh: A Four-Round DHS Machine-Learning Study
, M. Monimul Huq , M. Sabiruzzaman , Md. Rezaul Karim
The revised arXiv preprint is now on v2 and reflects the current BMC Public Health submission. In a four-round BDHS benchmark, validation-regime choice changed screening workload estimates more than architecture choice, making temporal splits and capacity-based metrics essential before programmatic use.
Beyond Sustainable Development Goal (SDG) Accelerators: Directed Evidence for Adaptive Sustainability Portfolios
, Md Jahid Hasan Imran , Md. Naim Molla , Luknath Debnath , Tonmoy Shil , Ehsanul Bashar Pranto , Md Mostafizur Rahman Likhon , Md Shafin Sanyan Saad , Md. Rezaul Karim
Across 114 countries, 84 directed SDG linkages survive false-discovery control, but no single goal is a stable universal accelerator. Policy should build adaptive portfolios on supported, time-lagged linkages instead of fixed rankings. Under review at Earth's Future.
The Depth Delusion: Why Transformers Should Be Wider, Not Deeper
, Md. Rezaul Karim
Width should grow 2.8× faster than depth in transformers, validated across 30 architectures up to 7B parameters. Past a critical depth, adding layers actively hurts, even though parameter count increases.
A Multi-Model Approach to English-Bangla Sentiment Classification of Government Mobile Banking App Reviews
Md. Naim Molla , , Md. Binyamin , Md Jahid Hasan Imran , Tonmoy Shil , Nura Rayhan , Md Rezaul Karim
Classical models outperform fine-tuned XLM-RoBERTa on 5,652 English-Bangla government banking app reviews; aspect analysis surfaces transaction speed and UI as top pain points; 16.1pp Bangla–English accuracy gap.