Application of Machine Learning to Predict Mental Health Disorders and Interpret Feature Importance

Abstract

The mental health and mental illness crisis has become increasingly acute in recent years, and many digital solutions with artificial intelligence at their core offer hope for reversing the deterioration of our mental health. Machine deep learning techniques can be used to analyse big data to build predictive models for psycho-education, assessment and screening to assess the mental health status of subjects, and can help the clinical community discover information that is not available to many traditional psychological research tools. This paper presents an in-depth analysis of a mental health survey and examines how it can be applied to the Al/ML domain of mental health research and how machine learning models can be used in this domain for fitting and prediction. Based on this, the importance of the presence or absence of current mental health disorders on other characteristics of respondents is assessed and visualised. It was found that the Cross Gradient Booster (Random Forest) model gave the best prediction fit among the various types of machine learning models, and the Grid Search algorithm was used to confirm that the final model had the highest accuracy of 0.79784 at a learning rate of 0.1. The Permutation Importance analysis revealed that the most important characteristic is whether or not the person has suffered from a mental health disorder in the past.