Publications

Long waiting times in hospital emergency rooms affect both patient satisfaction and the use of resources. This study builds a machine-learning framework to predict ER waiting times from 5,000 patient encounters, engineering features such as patient acuity, staffing ratios and time-of-day and seasonal effects. Linear models, tree-based ensembles and support vector machines were compared with 5-fold cross-validation on MAE, RMSE and R². A hyperparameter-tuned stacked ensemble performed best, reaching R² = 0.9473 — a validated, data-driven tool to help hospital administrators manage resources proactively and communicate better with patients.

Breast cancer is a significant global health concern, with increasing incidence and mortality rates. Current diagnostic methods face challenges, highlighting the need for improved approaches. By applying a variety of machine learning algorithms — conventional (KNN, SVM, ANN, RF, XGBoost) and ensemble models (AutoML) — as well as deep learning techniques, this work addresses the difficulties in breast cancer diagnosis. Synthetic data generation models are leveraged to enhance the analysis, comparing the efficiency and accuracy of these models on both original and synthetic datasets.

Supplier selection is a critical aspect of purchase management within the supply chain, particularly for a Bangladeshi jute company, involving the evaluation of both qualitative and quantitative criteria. This research integrates machine learning (ML), the fuzzy Analytic Hierarchy Process (AHP) and Bayesian Belief Networks (BBNs) into a user-friendly interface. Key selection criteria were identified through expert interviews and a literature review, then weighted with fuzzy AHP; BBNs assess supplier risk factors, incorporated into the ML model. The model achieves an accuracy of over 95% in selecting the most suitable suppliers based on user-defined requirements.

Cardiovascular disease (CVD) is a major global cause of death, and timely diagnosis is crucial — especially in rural areas of developing nations where heart specialists are scarce. This study compared ML classifiers for diagnosing CVD — SVM, Decision Tree, ANN, Logistic Regression, KNN, bagging variants, Gaussian Process and AutoML — on accuracy, precision, recall and F1 score, using the UCI Heart Disease data. AutoML emerged as the best classifier. A risk-factor analysis with a Bayesian Belief Network then assessed heart-disease risk among individuals predicted positive by AutoML.