Predicting Cancer Relapse with Machine Learning from an Open Brazilian Database

Currently, there is not an established knowledge about the correlation between the cancer type, the characteristics of the patients, the treatment, among other features of interest, with the occurrence of relapse. The goal of this work is to analyze the feasibility of applying machine learning techniques to predict the risk of relapse of several types of cancer. We aim to obtain a minimal number of features from the clinical records of an open public database to develop reliable prediction models. This database, from ONCOCENTRO Foundation of Sao Paulo State-Brazil, contains 905,556 registries of patients with 331 types of cancer. The best classification model obtained here showed, for a specific type of cancer, an accuracy of 83.08, +− 0.2 SD, and F1 Score mean of 0.83. This article also discusses possible improvements and future research involving the application of other machine learning techniques.