Computationally Efficient Explainable AI Framework for Skin Cancer Detection

Skin cancer stands among some of the fastest growing and fatal malignancies of the world as a result early and accurate diagnosis of skin cancer is essential in order to enhance patient survival and treatment prognosis. Conventional methods of diagnosis including dermoscopy and histopathological examinations are expensive and time consuming also subject to inter-observer error. To address these shortcomings, this research suggests an optimized hybrid cascading framework that combines deep and machine learning strategies towards building an AI model for the detection of skin cancer. This study applied five different Convolutional Neural Network (CNN) as well as seven Machine Learning (ML) model and three optimization technique. Among them the suggested MobileNetV2 + LDA + LR model had the best accuracy of 99.33%, precision of 99.47%, recall of 97.33%, and F1-score of 98.31%. Moreover, it was found that the framework had better computational efficiency, with a time complexity of 35.1 ± 1.24 seconds and moderate memory usage. Later on, Explainable AI (XAI) including Grad-CAM++ and LIME techniques were employed to validate the interpretability of the model and revealed the clinically relevant areas of lesions that affected predictions, which increased the level of transparency and credibility.