This research introduces an interpretable deep learning framework for diabetic foot ulcer (DFU) classification by integrating a custom CNN architecture with multi-optimizer training, Genetic Algorithm (GA)-based model selection, and Grad-CAM explainability. Unlike conventional approaches that rely on manually selected optimizers and black-box predictions, the proposed framework systematically evaluates seven optimizers and automatically selects the best-performing model using evolutionary optimization. The selected AdamW-based CNN achieved 95.24% accuracy and 98.97% ROC-AUC on an independent test set while providing visual explanations through Grad-CAM to highlight clinically relevant ulcer regions. The framework offers a lightweight, accurate, and transparent AI solution to support reliable DFU screening and clinical decision-making.
