A Hybrid Approach for Skin Cancer Detection and Classification using DenseNet121

Skin cancer is a serious problem that gets over-looked often times. Sometimes it is not feasible for the human eye to accurately diagnose the disease from imaging data, whenever a doctor undertakes a manual examining. Methods that utilize deep learning are being implemented frequently in modern times to address issues associated with our daily lives. Consequently, researchers propose a computerized framework that utilizes deep neural network techniques for diagnosing dermatological problems. In the proposed model, we have used hybrid CNN model based on DenseNet121 by fusing 2 algorithms namely best first and genetic search and analyse the performance to detect benign and malignant melanoma. The reduced feature set is used to classify the skin cancer. The final phase involves classifying the test features using Random For-est, Naive Bayes, and BayesNet classifiers. Our model received best accuracy of 96.5% with BayesNet.