Deep Learning-Based Automated Skin Cancer Detection Using Convolutional Neural Networks

Abstract

This study investigates the classification of skin cancer lesions utilizing Convolutional Neural Networks (CNNs) on the HAM10000 dataset. The collection includes photos of seven unique categories of skin lesions: melanocytic nevi, melanoma, benign keratosis-like lesions, basal cell carcinoma, actinic keratoses, vascular lesions, and dermatofibroma. Our main goals were to examine diagnostics efficacy over several lesion types and assess CNN performance in precisely categorizing dermatoscopic pictures. Label encoding and data balancing were used to ensure class representation. For regularization, we built sequential CNN models with convolutional, pooling, and dense layers and dropout and batch normalization. Transfer learning and Adam optimization were used in training. Our classification accuracy averages above 73.8%, demonstrating solid performance across lesion types. We tested the models' generalization ability on new data and found consistent performance. Confusion matrix analysis showed accurate categorization with low misclassification. This research improves early identification and intervention, improving patient outcomes and healthcare efficiency.