Automated Prediction of Heart Disease Using AutoML-H20 Across Diverse Datasets

Heart Disease (HD) remains one of the leading causes of mortality worldwide. While angiography is the gold standard for diagnosing HD, it is costly and complex, limiting its accessibility. Traditional machine learning (ML) models have been applied to HD diagnosis, but these methods often require extensive manual tuning and domain expertise for optimal algorithm selection, making the process time-consuming and prone to human error. This study addresses these challenges by developing an automated machine learning (AutoML) approach using the H2O framework to predict HD effectively. Five datasets were used to evaluate the model’s performance, including the Cardiovascular Disease (CVD) from Mendeley data , Framingham from Kaggle, Heart disease ,Statlog and Extention of Z-Alizadeh sani (EZAS) datasets from UCI Repository. Key metrics such as accuracy, recall, precision, and F1-score were assessed. Results indicate that the DRF model achieves the highest accuracy on the CVD dataset, while the GLM model performs best for the Framingham, Heart, and Statlog datasets. For the EZAS dataset, the GBM model shows the highest predictive accuracy. Overall, models perform exceptionally well on the CVD and EZAS datasets, achieving near-perfect scores, while the Framingham dataset exhibits comparatively lower performance. This study demonstrates the potential of AutoML to streamline HD diagnosis, offering a more accessible and efficient alternative to traditional ML methods.