Worldwide, coronary artery disease (CAD) ranks high among the main killers and disablers. It is critical to identify high-risk patients for coronary artery disease (CAD) early and administer preventative treatments to enhance patient outcomes. Algorithms for machine learning (ML) have demonstrated potential in CAD risk prediction but their performance can vary depending on the input features and modeling approach used. In this study, we aimed to identify the key predictors of CAD using a combination of statistical methods and novel ML algorithms, and compare the predictive accuracy of different modeling approaches. We analyzed data from 1000 patients, including 500 with confirmed CAD and 500 controls. Predictor variables encompassed demographic, clinical, imaging, and genetic factors. After identifying the most important predictors by univariate and multivariate logistic regression, we chose the best feature subset using recursive feature elimination. Next, we used stratified tenfold cross-validation to train and assess a number of machine learning models, such as deep neural networks, logistic regression, random forest, and gradient boosting. Age, sex, diabetes, hypertension, dyslipidemia, smoking, family history, C-reactive protein, and polygenic risk score were the most significant predictors of CAD (all p < 0.001). With an area under the receiver operating characteristic curve (AUC) of 0.89 (95% CI: 0.87–0.91), the top-performing machine learning model outperformed the standard logistic regression model (AUC 0.82; p < 0.001). Feature importance analysis revealed that the ML models were able to capture complex non-linear interactions between predictors. Overall, our findings demonstrate the utility of combining statistical and ML approaches to identify the key drivers of CAD risk and develop accurate prediction models. With further validation, such models could potentially be deployed in clinical settings to guide personalized prevention and treatment strategies.