A rigorous comparative study of three supervised machine learning classification algorithms — K-Nearest Neighbours (K-NN), Decision Tree, and Random Forest — applied to two real-world inspired classification problems: (i) predicting school dropout among students in India, and (ii) diagnosing the presence of heart disease using clinical parameters from the UCI Heart Disease dataset.
The study covers the complete machine learning pipeline: dataset acquisition and preparation, feature engineering, model training, hyperparameter configuration, and comprehensive performance evaluation. Evaluation metrics include Confusion Matrix, ROC-AUC curve, Validation Accuracy, Log-Loss, and Misclassification Analysis.