Cardiovascular disease remains a leading cause of global mortality, underscoring the need for robust early detection models. This study introduces a novel hybrid approach combining Earthquake Dynamics-Based Optimization (EDBO), Modified Adaptive Differential Evolution (MadDE), and Random Forest (RF) for enhanced heart disease prediction. The proposed model leverages EDBO’s global feature exploration capabilities alongside MadDE’s local refinement strength, optimizing feature selection and improving predictive performance. A logistic growth mechanism is integrated to simulate realistic population dynamics, accelerating convergence during optimization. To mitigate overfitting from high-dimensional data, EDBO reduces the feature space, while MadDE’s even/odd iteration strategy ensures a balanced exploration-exploitation trade-off for refined local optima. Rigorous cross-validation confirms the model’s robustness across three datasets: the Framingham Heart Study, Cleveland Heart Disease, and a self-collected clinical dataset. Experimental results demonstrate significant performance gains over baseline RF, with accuracies of 90.15% (Cleveland), 84.3% (Framingham), and 98.8% (clinical dataset)—surpassing traditional RF’s 83.6%, 83.74%, and 93.25%, respectively. The hybrid model also achieves superior precision, recall, and F1 scores. Notably, only 10 critical features are selected from the clinical dataset, reducing computational overhead while enhancing performance.