Abstract:
Since cardiovascular disease (CVD) is the leading cause of death worldwide, there is an urgent need for early and accurate diagnostic tools. A novel and potent method for building such highprecision predictive models from intricate clinical data is machine learning (ML). Based on an optimized ensemble learning strategy, this thesis proposes a strong methodological framework for cardiovascular disease prediction. Instead of focusing on a single algorithm, our study assessed and directly compared a number of top models. We started by methodically finetuning three potent "base-learners": Random Forest (RF), XGBoost, and LightGBM. To capitalize on their combined strengths, these were then used to build two sophisticated ensemble models: a soft-Voting Classifier and a Stacking Classifier. A complete set of diagnostic metrics, including accuracy, precision, recall, and F1-score, were used to rigorously evaluate each model on a held-out test set. The outcomes were conclusive. With a balanced F1-score of 0.97 and an accuracy of 96.74%, our Stacking Classifier was the best. This analysis highlights the model's true potential as a dependable tool for clinical decision support by confirming its high sensitivity (Recall) and positive predictive value (Precision).
