Identification and validation of key predictive factors for heart attack diagnosis using machine learning and fuzzy clustering

Heart disease remains a leading cause of mortality worldwide, making early diagnosis and prevention critical for improving patient outcomes. However, accurate diagnosis often depends on physician expertise, which can lead to medical errors. This study develops a machine learning-based predictive model to assist physicians in diagnosing and predicting heart attacks with higher precision. Patient data from an Iranian hospital were collected and preprocessed, including handling missing values, normalization, and outlier removal. Analysis using RapidMiner software identified seven significant predictors of heart attack from thirteen initial factors, achieving a high accuracy of 97.86%, confirmed by hospital physicians. Among various machine learning algorithms, the decision tree demonstrated the highest accuracy (97%) and provided interpretable decision pathways for clinical use. Decision rules extracted from the decision tree highlighted chest pain, trait-anxiety, age, smoking, and resting electrocardiographic results as the most influential factors. To validate these findings, fuzzy clustering was employed, revealing an alignment rate of 76% for the two most critical factors-chest pain and trait-anxiety. These findings emphasize the prioritization of these factors during initial evaluations, followed by age, smoking, and resting electrocardiographic results to enhance diagnostic accuracy. The integration of machine learning and fuzzy clustering offers a robust approach to clinical decision-making, verified by expert feedback from hospital physicians, and contributes to reducing diagnostic errors.