Research on Skin Cancer Classification Based on an Optimized AlexNet Model

Abstract

Skin cancer is one of the most common types of tumors worldwide, and its early detection and diagnosis are essential for enhancing treatment efficacy and patient prognosis. Nevertheless, conventional pathological examination techniques are frequently time-intensive and inefficient, thereby constraining their applicability in large-scale clinical screening. To overcome these limitations, the current study introduces an enhanced AlexNet model grounded in deep learning, incorporating a dynamic learning rate adjustment strategy to refine the training process. By extracting discriminative features from images of benign and malignant skin cancers, the model was trained and optimized to attain superior performance. The finalized model demonstrated a classification accuracy of 85% on the skin cancer test dataset sourced from the Kaggle dataset, alongside an area under the ROC curve (AUC) of 0.9321, reflecting a robust capability to distinguish between benign and malignant lesions. The optimized AlexNet-based model developed in this study effectively achieves the classification of benign and malignant skin lesions, offering meaningful support for clinical diagnosis and decision-making.