Combined deep learning and machine learning models for the prediction of stages of melanoma

Melanoma is deadly kind of skin cancer as it gets metastasized soon. It is really essential to recognize melanoma and start treatment at early stage. It is also necessary to determine the stage of melanoma in order to treat melanoma patients. Non-invasive technique is required to detect the stage of melanoma. Proposed system presents novel technique to classify the stages of melanoma based on thickness of tumor. This system uses dimensionality reduction technique to reduce the number of features and it also uses combine approach of deep learning (DL) and machine learning (ML) algorithm which include multilayer perception (MLP) and random forest (RF). Deep learning method is always better for training as they can reduce the need for data preprocessing and feature engineering and can provide simple trainable models built using only five or six different operations. Secondly, they are scalable, as they can be easily parallelized on GPUs or TPUs and can be trained by iterating over small batches of data. Thirdly they are reusable, so they can be trained on additional data without starting from scratch, making them viable for continuous online learning. For classification task machine learning algorithm that is, random forest is used as it decreases over fitting in decision trees and aids to increase the accuracy. Total of three algorithms were used, MLP, RF and proposed algorithm combined multilayer perception and random forest that is, MLP-RF. Among these models, the MLP-RF showed the best results in predicting melanoma stages with the accuracy of 97.42%.