A Comparison of Methods for Predicting Heart Disease: Neural Network, XGBoost, Gradient Boost, Logistic Regression, and SVM

Cardiovascular diseases (CVD) are a major cause of mortality, demanding immediate attention. Early screening for risk factors speeds up CVD identification and treatment, reducing mortality risk efficiently. This study compares the predictive abilities of XGBoost, Gradient Boost, Neural Network, Logistic Regression, and SVM in forecasting CVD symptoms using CDC data from Kaggle. Evaluating against a heart disease risk approach, we assess model performance with AUC and $F_{1}$ - score for class 1 due to data imbalance. Additionally, we consider training run time as another metric for evaluation. XGBoost, Gradient Boost, and Logistic Regression excel in AUC and runtime, outperforming Neural Network and SVM significantly. The performances are quite similar when comparing the $F_{1}$ - scores, but Logistic Regression, XGBoost, and Gradient Boost still maintain their superiority. The paper recommends using Logistic Regression and boosting-based decision tree models for future heart disease prediction investigations.