Advanced Skin Cancer Detection using Hybrid CNN Feature Extraction

Abstract-Skin cancer is one of the deadliest types of cancer, with a rapidly increasing incidence worldwide. Early detection is crucial to reducing the mortality rate. In this paper, we present an effective computer-aided diagnostic model for accurate skin cancer detection and classification. Our proposed system consists of three primary steps: a) Preprocessing, b) Feature extraction, and c) Classification. During preprocessing, image quality is enhanced through median filtering. In the feature extraction phase, features are extracted from three powerful pretrained CNN models—GoogleNet, AlexNet, and ResNet-101—using transfer learning and are then combined. In the classification stage, the hybrid features are classified using three successful Machine Learning (ML) classifiers: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbor (KNN). We validated our model on 3000 images from the MNIST dataset, achieving an accuracy of 96.66%, a precision of 96.5%, a recall of 96.66%, and an F1-score of 96.5%.