Cardiovascular disease remains the leading cause of mortality worldwide, highlighting the need for accurate and early risk prediction to support effective clinical intervention. This study presents a customized Random Forest classifier for ten-year coronary heart disease prediction using the Framingham Heart Study dataset. To address the class imbalance inherent in the data, classweighted learning is employed while deliberately avoiding synthetic data generation to maintain the original data distribution. The proposed model achieved an overall accuracy, precision, recall, and F1-score of 81%, with a ROC–AUC of 72%, indicating moderate discriminative ability. SHAP-based explainability analysis identified age and systolic blood pressure as the most influential predictors, followed by sex, daily cigarette consumption, and a history of hypertension, aligning with established cardiovascular risk factors. However, class-wise evaluation revealed limited sensitivity for the positive class, with a recall of 34% and a precision of 37%, reflecting the challenges posed by the severe class imbalance in the Framingham dataset. In contrast, the model demonstrated strong performance in identifying negative cases, achieving 88% precision and 90% recall. By integrating predictive modelling with clinical interpretability, this study demonstrates the applicability of interpretable machine learning for healthcare settings where transparency and clinical relevance are essential.