Performance comparison of machine learning algorithms in heart disease prediction: a case study from Türkiye

In recent years, rapid advancements in communication and information technologies have transformed big data analytics into a strategic component of healthcare services; these developments have enabled the more systematic use of machine learning and artificial intelligence–based approaches in early disease diagnosis and risk classification. Machine learning algorithms can extract meaningful patterns from multidimensional and complex data structures—such as genetic, lifestyle, clinical, and demographic indicators—and thereby achieve high performance in determining disease risk. Although the literature shows a strong concentration of studies conducted on prepared data set, standardized datasets, such datasets do not always fully reflect the diversity and heterogeneity of real clinical conditions. In this context, studies that methodologically align with research using prepared datasets but provide an original contribution to the literature by utilizing local and real patient data have gained increasing importance. This study aimed to determine the most successful machine learning model for predicting heart disease risk using a dataset obtained from the electronic medical records of patients who visited the cardiology clinic of a private hospital in Denizli between 2019 and 2024. Four different scenarios were defined to examine the effect of missing data on model performance, and various missing data handling strategies were applied. The resulting models were comprehensively evaluated in terms of performance metrics such as accuracy, precision, sensitivity, and F1 score. The findings reveal that machine learning models based on real clinical data can contribute to decision support mechanisms in cardiovascular risk prediction, thereby strengthening early diagnosis processes and supporting personalized treatment planning.