Performance Evaluation of Supervised Machine Learning Models for Heart Disease Prediction Using Clinical Survey Data

This study focuses on evaluating the performance of supervised machine learning models using clinical survey data. We constructed a machine learning pipeline to evaluate the performance of the heart disease prediction models. For this study, we collected clinical data from 399 patients at Cumilla Medical College Hospital. The features included gender, age, blood group, smoking, diastolic pressure, systolic pressure, glucose, hypertension, chest pain (CP), BMI, family medical history (family P), stroke, and heart disease (Heart problem). Heart disease (Heart Problem) was used as the target variable. Initially, we encoded all categorical features, including the binary target label, then separated the target variable. We used Random Over-Sampling to balance the distribution of classes. We applied random over-sampling only on the training dataset. We randomly divided the dataset into 80% training and 20% testing sets. We trained and evaluated five machine learning models on split data, including SVM, Random Forest, Logistic Regression, Decision Trees, and K-Nearest Neighbors. Finally, we calculated accuracy, precision, recall, and the F1-score from the Confusion matrix analysis and 5-fold cross-validation of these models. Then, we compared the evaluation performance of these models. We observed that the Random Forest achieved the highest test accuracy of 90%, and the cross-validation accuracy was 93%. The Support Vector Machine and Logistic Regression achieved slightly lower accuracy. K-Nearest Neighbors had the poorest performance, comparatively lower recall for the positive class. Ensemble models, notably Random Forest, appear to have potential in clinical practice but should be evaluated in larger and more heterogeneous populations. Using clinical data that was acquired locally, this work offers a comparative performance benchmark and highlights generalization behavior across validation methodologies.