A Novel Approach to Skin Cancer Detection using CNN, ResNet, and Hybrid Feature Extraction

Skin cancer is a prevalent and potentially life-threatening condition, making early and accurate detection crucial for effective treatment. This work proposes a novel hybrid approach to skin cancer detection, integrating both deep learning and traditional ML techniques. Utilizing a dataset from Kaggle, the process began with initial classification using a Convolutional Neural Network (CNN) and achieved an accuracy of 90%. Features were then extracted using Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and ResNet, adding multiple layers of feature richness. These features were combined into a comprehensive hybrid feature vector and analyzed using ML models, including Random Forest, XGBoost, and AdaBoost. The Random Forest model achieved an accuracy of 96%, while XGBoost obtained an accuracy of 95. 60%. AdaBoost reached an accuracy of 95%. The proposed method demonstrated significant improvements in detection accuracy and robustness, highlighting its effectiveness in enhancing skin cancer detection.