Assessing the Robustness of Machine Learning Algorithms for Cardiovascular Disease Detection Across Diverse Clinical Datasets

Cardiovascular diseases (CVD) continue to pose significant health risks globally, accentuating the need for early and precise detection mechanisms. With the evolution of computational methods in healthcare, machine learning offers transformative solutions for diagnostic accuracy. This research aims to identify an algorithm with consistent performance across multiple datasets for potential integration into a cardiac disease prediction platform. We examined nine prominent machine learning algorithms, namely Support Vector Machine (SVM), Gradient Boosting (GB), Random Forest (RF), Logis-tic Regression (LR), Decision Tree (DT), K-Nearest Neigh-bor (KNN), Naive Bayes (NB), Extreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP), and evalu-ated their predictive performance across two heterogeneous datasets. Both datasets encompass 14 attributes but differ in instance sizes: 303 and 1025, respectively. Through a meticulous methodological framework, the data underwent preprocessing, splitting, and model training, followed by validation using metrics such as Precision, Recall, F1 score, and Accuracy, coupled with a confusion matrix for detailed class-based evaluation. Our findings revealed that the Random Forest and MLP algorithms exhibited superior consistency and robustness in disease prediction across both datasets, achieving a peak accuracy of 95.14%. While XGBoost performed proficiently on one dataset, its performance wavered in a cross-dataset scenario. Based on these findings, either the Random Forest or Multilayer Perceptron models are recommended for developing a robust heart disease prediction system. This research not only affirms the potential of machine learning in revolutionizing CVD diagnostics but also underscores the importance of algorithm selection based on dataset characteristics.