The rising global incidence of skin cancer underscores the critical need for diagnostic systems that combine high accuracy with interpretability to build trust among clinicians and patients. Deep learning (DL) has revolutionized medical imaging, achieving unprecedented accuracy in skin cancer diagnosis. However, the inherent “opaque” nature of these models limits interpretability, which erodes trust and hinders clinical adoption. To address these challenges, we propose a novel interpretable DL framework that combines an auto-encoder for efficient feature extraction and a CNN for accurate classification. To enhance trust and usability, we integrate post-hoc explainability techniques, such as Gradient-weighted class activation maps and saliency maps, to visualize model decisions and align them with dermatological reasoning. Our framework uniquely emphasizes detailed regional analysis, empowering clinicians and patients to validate AI-driven decisions with confidence. By integrating high accuracy with transparency, our framework enhances diagnostic workflows and sets a precedent for ethical scalable AI deployments in dermatology and beyond.