<p>Cardiovascular diseases (CVD) remain the leading cause of global mortality, underscoring the need for predictive models that are both accurate and interpretable. Existing approaches often struggle with overfitting, limited transparency, and inconsistent performance in clinical settings. To address these challenges, we propose ClassifyIT, a modular classification pipeline for early CVD prediction. The pipeline integrates three components: (i) MIST-CC, a feature selection method based on mutual information and spanning trees that reduces redundancy while preserving interpretability; (ii) STIR, an input regularization strategy that structures feature flow into neural networks to mitigate overfitting; and (iii) IPANN, a novel deep learning architecture that propagates input features across layers to capture non-linear interactions. Evaluation was performed on the Cleveland Heart Disease dataset using 10-fold cross-validation. ClassifyIT achieved an accuracy of 87.16%, outperforming conventional deep neural networks (78.80%) and classical classifiers such as logistic regression and SVM. Ablation studies confirmed the incremental contributions of each component, while error analysis highlighted challenges in borderline clinical cases. With its robust performance, interpretability, and modular design, ClassifyIT offers a promising decision support tool for early disease detection and can be adapted for broader healthcare applications.</p>