Predictive Modeling of Coronary Artery Disease Using Machine Learning

CAD remains among the leading causes underlying morbidity and mortality worldwide, requiring a prompt and precise diagnosis to improve outcomes for patients. Using the Z-Alizadeh Sani dataset, which consists of 303 samples with 55 features categorized into demographic, symptoms and examination, electrocardiogram, laboratory, and echo features, this study investigates the application of machine learning models for the prediction of CAD. To guarantee data quality, preprocessing procedures included outlier detection, univariate analysis, and a check for missing and duplicate values. Accuracy, precision, recall, F1 score, ROC AUC score, and an ensemble approach are used to gauge the performance of the models under evaluation, which include CatBoost, XGBoost, LightGBM, and an ensemble approach. The class imbalance was addressed by using SMOTE which improved the recall of CAD predictions. With a perfect recall rate of 100%, an F1 score of 0.945, and an accuracy of 91.8%, CatBoost outperformed the other models. One important factor in medical diagnostics is false negatives, which can be reduced with a high recall rate. Using Optuna, hyperparameter tuning was done to optimize the model parameters for better predictive performance. The study highlights that in order to develop trustworthy diagnostic tools for the medical field, sophisticated machine learning techniques and appropriate data handling are essential. Subsequent research endeavors will center on the explainability of the models, their amalgamation with clinical data, and their validation throughout various datasets to bolster the resilience and relevance of the predictive models.