Accurate Melanoma Detection Using XceptionNet and DenseNet121: A Deep Learning Framework

Melanoma is among the most lethal skin cancers, where early detection is critical for improving patient outcomes. While deep learning has shown promise in dermatological image analysis, existing models often struggle with generalization, class imbalance, and interpretability. This study presents a novel dualmodel deep learning framework that combines XceptionNet and DenseNet121 through late fusion to enhance melanoma classification. The proposed approach leverages the strengths of both architectures-XceptionNet's efficient spatial feature extraction and DenseNet121's dense feature reuse-to improve diagnostic accuracy. To address dataset imbalance, in the proposed work, weighted cross-entropy loss and data augmentation techniques are employed. Evaluated on the ISIC 2020 dataset, this model achieves 94.7 % accuracy, outperforming individual networks (XceptionNet: 91.2%, DenseNet121: 92.4%) and demonstrating superior sensitivity (93.6 %) and specificity (95.9 %). These results indicate that the fusion strategy effectively captures complementary features, reducing misclassification risks. The framework's high sensitivity is particularly valuable in clinical settings, where missing malignant cases can have severe consequences. This research contributes to AI-assisted dermatology by providing a more robust and interpretable classification system. Future work will focus on integrating explainability mechanisms and validating the model across diverse populations to ensure broader clinical applicability.