Enhancing the Prognosis of Heart Disease via Hyperparameter Optimisation of Machine Learning Models and Outlier Detection

Abstract

Information about medical research has verified that the most common cause of death and health loss nationwide is heart disease, which affects the heart and blood vessels. The number of premature deaths could be reduced by early detection of certain illnesses. Machine learning and data mining continue to be leveraged in identifying diseases based on an individual's distinct traits. The complexity of understanding the datasets' goals, the presence of too many variables to examine, and the lack of performance accuracy have, nevertheless, frequently posed difficulties for these approaches. This study presents a machine learning (ML) based approach for cardiovascular disease (CVD) prediction using integrated datasets (Cleveland and Statlog) to enhance model robustness. The proposed work addresses key challenges such as class imbalance and outlier influence, which often degrade predictive accuracy in clinical data analysis. Data preprocessing techniques, including balancing methods and outlier removal, were applied to ensure a reliable dataset. Three supervised ML classifiers, Random Forest (RF), Gradient Boosting (GB), and Support Vector Machine (SVM), were trained and evaluated using five-fold cross-validation and RandomizedSearchCV for hyperparameter optimisation. Performance was measured employing F1-score metrics, recall, accuracy, and precision. Among the models, RF demonstrated the highest predictive accuracy of 98.2%, followed by GB at 97.6%, indicating the effectiveness of ensemble methods in heart disease prediction. The findings highlight machine learning’s potential for clinical decision assistance and risk assessment in cardiovascular health.