Modelling the risk ecosystem of depression using machine learning in a population of young adults

Abstract

AbstractUnderstanding what is predictive of early adulthood depression could help inform resource targeting and direction of approaches aiming to alleviate the personal, cultural, and economic burden of depression and similar disorders. This work uses multivariate longitudinal data (n=3487) measured from conception to adulthood from a UK based birth cohort of young adults (Avon Longitudinal Study of Parents and Children (ALSPAC)) and a machine learning approach to a) investigate whether episodes of early adulthood depression can be predicted from various risk factors across early life and adolescence, and b) interpret which factors are most important for predicting episodes of early adulthood depression. Here, we build four models to predict participants having an episode of early adulthood depression and show that the highest performing model can predict if people experienced symptoms of depression with an F1-score of 0.66, using a range of biological, behavioural, and early life experience related risk factors.