Towards Improved Skin Cancer Diagnosis: A Reliable Predictive Framework

Skin cancer, particularly melanoma, represents a major global health concern due to its high mortality rate and aggressive progression. Accurate and timely diagnosis is crucial for effective treatment, yet conventional diagnostic approaches remain subjective and prone to inconsistencies. This study presents a novel ensemble deep learning framework designed to improve the accuracy and reliability of automated skin cancer classification. The proposed method combines the strengths of InceptionV3 and MobileNetV2 architectures to extract both global and local image features, enabling robust lesion differentiation. Leveraging the ISIC 2018 dermoscopic image dataset encompassing seven distinct skin lesion categories, the framework incorporates targeted data augmentation techniques to mitigate class imbalance. Experimental evaluation demonstrates that the model achieves a high classification accuracy of 95%, surpassing leading models such as EfficientNetB4 (87.91%) and DarkNet-53 (85.4%). The model also demonstrates strong performance metrics with an average precision, recall, and F1-score of 0.94, 0.94, and 0.93, respectively, validating its robustness and generalizability. While the framework shows promising potential for real-world clinical and telemedicine deployment, future research will focus on enhancing interpretability through explainable AI, expanding dataset diversity, and optimizing computational efficiency via model compression.