Skin Cancer Classification: A Comparative Analysis of MobileNet and Hybrid Model (Xception+Efficient Net+GCN)

Skin cancer, caused by the abnormal growth of skin cells due to prolonged exposure to UV light, includes common forms such as squamous cell carcinoma, basal cell carcinoma, and melanoma. Early and accurate detection is crucial for effective treatment. This study compares the performance of a newly developed hybrid model with several well-known pre-trained models, including MobileNet and EfficientNet, for skin cancer image classification. The analysis focuses on popular convolutional neural network (CNN) architectures, as well as a hybrid model that combines Xception, EfficientNet, and Graph Convolutional Networks (GCN) to improve feature extraction and classification accuracy. All models were fine-tuned on the Skin Cancer ISIC 2020 dataset from Kaggle, which consists of 11,396 images— 9,200 used for training (5,200 benign and 4,000 malignant) and 2,196 for testing (1,100 benign and 1,096 malignant). Among developed models, the MobileNet model achieved an accuracy of 75% with a runtime of 40 seconds, the other models substantially shown little promising results, while EfficientNet achieved an accuracy of 77% with a runtime of 43 seconds. The hybrid model, however, significantly outperformed both, achieving 90% classification accuracy with greater efficiency. In addition, a user-friendly interface was developed using the Flask framework, allowing users to easily test and validate the model's performance.