Enhanced skin cancer classification using a combination of bag of words and deep Q-network features with ReliefF feature selection

Skin cancer represents a major worldwide health challenge, and timely detection is crucial to ensure proper treatment of this malignant condition. Traditional diagnostic techniques, such as visual inspection and biopsy, frequently require time, skill, and medical resources. In this study, a hybrid feature extraction method using bag of words (BoWs) feature function and Deep Q-Network (DQN) are proposed with the ReliefF feature selection to classify multiple types of cancer skin. Preprocessing, feature extraction, feature selection, and classifier are the four phases of the proposed model. The combination of BoWs’ feature extraction with DQN’s deep feature extraction yields that BoWs captures texture, color, and spatial patterns, while DQN captures global context by encoding complex visual cues and lesion morphology that BoWs is unable to capture. ReliefF selects only the most relevant features from the combined BoWs and DQN-derived features, ranking features according to their capacity to differentiate between classes. The publicly available ISCI-2019 dataset was used in this study. Overall, the results showed how the hybrid feature extraction model improved accuracy in the classification of skin cancer. Future research challenges involve enhancing pre-processing techniques and applying optimization algorithms for tuning DQN and classifier hyperparameters, which are crucial for improving the performance of the proposed approach.