The main challenge for health insurers is to move from reactive to proactive management of their clients, that is, prioritize prevention and implementation of actions that favor health in its physical, mental, and social dimensions, and not only that mitigate the disease. This article presents a systematic review of the literature on the main machine learning methodologies that allow, through the prediction of mental illnesses, to carry out an early intervention. It was found that the main methodologies for this purpose are statistical models such as logistic regression and neural networks; and that the different indicators of neuroimaging and behavior in social networks become fundamental predictor variables when it comes to predicting mental illnesses. © 2021, Associacao Iberica de Sistemas e Tecnologias de Informacao. All rights reserved.