This paper presents a transparent and interpretable machine learning pipeline for heart disease prediction based on structured clinical data. The proposed methodology emphasizes three essential properties for medical AI systems: interpretability, robustness to outliers, and generalizability. Multiple supervised learning models were developed and rigorously evaluated, including Logistic Regression, Random Forest, and Decision Trees. While Random Forest achieved strong predictive accuracy, its black-box nature limited its transparency. Logistic Regression offered better interpretability but required advanced preprocessing techniques such as winsorization, feature standardization, and polynomial expansion. Ultimately, a Decision Tree classifier—optimized through pre-pruning strategies (e.g., max_depth = 10) and using the Gini impurity criterion—was selected for its optimal trade-off between performance and clinical comprehensibility. The resulting pipeline provides a practical and trustworthy foundation for developing interpretable and deployable AI solutions in healthcare contexts.