Deep learning models for predicting heart disease risk using the UCI database: methods, performance, and clinical context

Abstract

Objectives: To develop and evaluate deep learning models for predicting heart disease usingthe University of California, Irvine (UCI) heart disease dataset, and to contextualize model performance against classical machine learning approaches. Method: Data were extracted from the University of California Irvine (UCI) heart disease data-set, including information from Cleveland, Hungary, Switzerland, and Long Beach V, collected in 1988. The dataset comprises 1,025 patients and 14 key attributes. Deep learning models were used to analyze the data and predict heart disease risk. Results: The deep learning models demonstrated high accuracy in predicting heart disease risk. The Random Forest model achieved an accuracy of 99. Significant predictors included exercise-induced angina and downsloping ST segments. The data revealed that 72 of females and 42 of males experienced heart attacks. There was a 79 chance that atypical angina and a 77 chance that non-anginal pain would lead to a heart attack. Exercise-induced angina had a 67 chance of resulting in a heart attack, while downsloping of the peak exercise ST segment had a 72 chance. Additionally, a 71 chance was observed for heart attacks in patients with no major coronary artery blockage (ca=0), and a 75 chance for those with a potentially reversible thalassemiarelated defect (thal=2). Age groups 40-44 and 50-54 had a 76 and 61 risk of heart attacks, respectively. Conclusion: Deep learning models can significantly enhance heart disease risk prediction, leading to improved treatment strategies. These findings can aid in early diagnosis and timely interventions, improving clinical outcomes for heart disease patients.