Among metabolic disorders, Type 2 Diabetes Mellitus (T2DM) stands as one of the most critical public health concerns of the twenty-first century. Its progression is insidious — damaging cardiovascular, renal,ocular, and neuralsystems long beforea formal diagnosis is reached. Building on the foundational framework explained in article [1], this studyconstructs an intelligent, multi-algorithm machine learning pipeline for early T2DM risk stratification using an identical clinical dataset of 61 patient records comprising rich biochemical markers: HbA1c, HOMA-R, HOMA-B, NEFA, hsCRP, Triglycerides, Cholesterol, HDL, LDL, VLDL, Serum Creatinine, and Insulin. Five classification algorithms — Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), Long Short-Term Memory (LSTM), and Hoeffding Tree (HT) — are comparatively evaluated against one another, with K-Means clustering applied as an unsupervised preprocessing enrichment step. Models are assessed across Accuracy, Precision, Recall, F1-Score, and AUC-ROC metrics. Results confirm that Random Forest and Hoeffding Tree achieve the highest classification accuracy of 91.67% and 91.53% respectively, with Random Forest additionally attaining a perfect Recall of 1.0 and an F1-Score of 0.93 — firmly establishing ensemble-based learning as the most reliable paradigm for clinical T2DM screening on small, feature-rich clinical datasets.
Loading publication timeline...
