Deep Learning Meets Traditional AI: A Hybrid CNN-RF Model for Accurate and Explainable Skin Lesion Classification

In dermatology, the task of skin lesion classification is a very important one, for early detection and treatment of skin cancer. In this work, we propose a hybrid AI model where Convolutional Neural Network (CNN) is used for feature extraction and a Random Forest (RF) classifier for final decision making. We evaluate the proposed model on the ISIC dataset: images of multiple types of skin lesion (melanoma, benign nevi, seborrheic keratosis). Experimental results show that the hybrid CNN_RF model surpasses stand alone models in accuracy, precision, recall, and F1 score, yielding an accuracy of 89.5%, precision of 88.9%, recall of 89.2% and F1 score of 89.0%; the standalone CNN and RF models resulted in accuracies of 85.3%, 78.6% respectively. The method proposed successfully strikes a good balance between feature learning and classification robustness, using the resultant model to provide better generalization. Finally, we perform qualitative analysis using Grad-CAM visualizations to motivate the model towards focusing on relevant lesion regions which improves interpretability. Moreover, the RF classifier also provides the feature importance insight to support clinical decision making. Receiver Operating Characteristic (ROC) curve analysis, confirms that the model has a strong discriminative power with an Area Under the Curve (AUC) of 0.92. The hybrid model shows stable learning without sharply decreasing the training set error; in other words, there is no strong overfitting of training data. The learning process also demonstrates stable training and validation accuracy curves and can be used in real applications. The results show that the CNN RF hybrid model is a promising solution for automated skin lesion classification which provides higher accuracy, interpretability and robustness.