Data-Driven Insights into Mental Health: Mapping Multi-Level Predictors to Longitudinal Outcomes

A key challenge in predicting a person’s internal state of mind is that there are a wide range of contributing factors that each have a subtle, yet significant, influence on mental health. To address this challenge, we trained machine learning algorithms on multiple sources of variation that could meaningfully contribute to psychological distress. Data mining techniques were used to identify key risk factors for predicting current symptoms and longitudinal outcomes from the Adolescent Brain Cognitive Developmental dataset (n = 11,552). Our results consistently revealed that social conflicts were the strongest indicators of psychopathology. Family fighting was the best predictor of current symptoms, whereas social victimization among peers was the top predictor of changes across time and future symptom severity. Sex-differences also emerged as a critical factor for predicting psychopathology, as females exhibited greater symptoms on average, and their symptoms became more severe over time. The long-term mental health of females was best predicted by social exclusion and reputational damage, whereas aggression and anti-social tendencies were the strongest predictors among males. While these findings provide novel insight into the developmental origins of psychopathology, our best performing models could only explain up to 40% of the variation between individuals. Future research is needed to obtain a more complete understanding of all the factors that meaningfully contribute to mental health.