Deep learning approach for automated skin cancer detection with comparative analysis of different batch sizes in dermatological image classification

Skin cancer is a significant public health issue, and accurate, efficient diagnostic methods are essential to reduce delays and errors associated with traditional screening, which often suffers from subjectivity and variability. This study aims to develop a reliable automated skin cancer detection system capable of improving diagnostic accuracy and reducing false negatives. We implemented an automated classification framework based on the Xception deep learning architecture. The system utilized pre-trained ImageNet weights for transfer learning, enabling effective feature extraction from dermoscopic images. Convolutional layers were used to capture discriminative features, which were combined with fully connected layers to perform binary classification of skin cancer versus not skin cancer. The proposed model achieved an overall accuracy of 99.34%, with a precision of 98.51%, recall of 98.80%, specificity of 99.52%, and an F1-score of 98.63% at a batch size of 64. These results demonstrate superior performance in early detection compared to traditional diagnostic methods. The Xception-based automated system offers a highly accurate, robust, and scalable solution for skin cancer detection. With its potential integration into dermatology clinics and telehealth platforms, the model can aid clinicians in achieving earlier and more reliable diagnoses, ultimately improving patient outcomes.