A procedural overview of why, when and how to use machine learning for psychiatry

Abstract

Machine learning (ML) is becoming a tool of choice to analyze high-dimensional datasets pertaining to mental health. Given the rapid integration of ML into research and clinical settings, this article provides a functional overview of a common ML pipeline used for the assessment and prediction of psychiatric disorders. Developing such a construct entails building a data infrastructure, collecting and preprocessing data, training and testing models and interpreting their results. Practical considerations pertaining to data management and preprocessing are first presented. We then describe considerations and best practices for model selection on the basis of the psychiatric disorder and the data modalities available for analysis. A critical analysis of existing works utilizing ML methods for psychiatric disorder assessment, prediction and causal associations is also provided. Last, future ML trends in psychiatry are highlighted. To reinforce learning, the Supplementary Note links to an interactive Jupyter Notebook that offers practical examples and hands-on interaction with a sample dataset. This Review provides a comprehensive overview of the principles, processes and procedures in the application of machine learning for psychiatry and mental health research.