Sudden cardiac arrest (SCA) is a life-threatening clinical event most commonly triggered by myocarditis, ischemic heart disease, a spectrum of cardiomyopathies, and primary electrical disorders. Major challenges in SCA research persist due to diverse underlying conditions and constraints in data collection processes. Recent advances in machine learning have demonstrated the potential for automated SCA risk stratification; however, dominant approaches often rely on single-modality inputs and opaque deep learning models, thereby limiting interpretability and external validity. This study proposes an interpretable multimodal learning framework for SCA prediction that integrates cardiac magnetic resonance imaging (MRI) images, and blood biomarker measurements obtained from publicly available databases. MRI signal is processed using deep convolutional neural networks (CNNs) to learn latent feature representations, while structured biomarker features are modeled using conventional machine learning classifiers. To ensure robustness and explainability across modalities, multiple algorithms, including Support Vector Machines (SVM), Random Forests (RF), and Gradient Boosting Machines (GBM), are evaluated. Results demonstrate that multimodal ensembling significantly improves predictive accuracy and recall compared to unimodal baselines, and that tree-based models provide clinically interpretable decision rules suitable for practical deployment.