Multi-Model Hybrid ML-DL Framework for Early Detection of Skin Cancer

One of the most deadly forms of cancer in the current scenario is skin cancer. As it is a rapidly growing global health concern, thus early detection is important to lower the death rates. This study seeks to enhance the diagnostic accuracy and dependability, as a hybrid approach of machine learning (ML) and deep learning (DL) is used. The proposed model combines conventional machine learning approaches like support vector machines (SVM), K-Nearest Neighbor (KNN), Random Forest (rf) and XGBoost. In this approach different types of Convolutional Neural Network (CNN) models like ResNet50, EfficientNet-B0 and Inception-V3 are used for extracting features from the dermoscopic images. This also helps in resolving the classification of different skin lesion types. The hybrid approach makes use of deep learning models for managing the complex patterns such as lines, shapes, textures,etc. and classical models for classification into different skin lesion types. The hybrid model when compared to deep learning or classical models individually, it provides an accuracy of 95.92%, sensitivity of 95.92%, precision of 95.96%, and F1-Score of 95.86%. The obtained results are based on large-scale testing that has been done on the dataset of skin lesions publicly available on kaggle.