Early accurate diagnosis of cardiovascular diseases (CVDs) has become important considering their increasing prevalence and high mortality rate. This paper presents an innovative deep learning-based approach for heart disease prediction, employing two ensemble frameworks: EnsCVDD-Net and BlCVDD-Net. Utilizing a dataset of 400,000 samples available on Kaggle, this research explores the combined effectiveness of LeNet and GRU models within the ensemble networks. Results demonstrate that both proposed models surpass traditional models, with EnsCVDD-Net achieved an accuracy of 90%, while BlCVDD-Net, blending LeNet, GRU, and MLP, also achieved 91% accuracy, outperforming both standalone LeNet and GRU models. EnsCVDD-Net exhibited optimal performance in the detection of CVDs, validated through various metrics such as accuracy, precision, recall, and F1-score. The proposed models utilize Adaptive Synthetic Sampling for data balancing and SHAP for model interpretability, providing insights into feature contributions. This proposal focuses on the effectiveness of the ensemble learning models in improving prediction accuracy and holds promise for practical applications in early cardiovascular disease detection.