Using ecological momentary assessment and machine learning techniques to predict depressive symptoms in emerging adults

Abstract

The objective of this study was to predict the level of depressive symptoms in emerging adults by analyzing sociodemographic variables, affect, and emotion regulation strategies. Participants were 33 emerging adults (M = 24.43; SD = 2.80; 56.3 % women). They were asked to assess their current emotional state (positive or negative affect), recent events that may relate to that state, and emotion regulation strategies through ecological momentary assessment. Participants were prompted randomly by an app 6 times per day between 10 am and 10 pm for a seven-day period. They answered 1233 of the 2058 surveys (beeps), collectively. The analysis of observations, using Machine Learning (ML) techniques, showed that the Random Forest algorithm yields significantly better predictions than other models. The algorithm used 13 out of the 36 variables adopted in the study. Furthermore, the study revealed that age, emotion of worried and a specific emotion regulation strategy related to social exchange were the most accurate predictors of severe depressive symptoms. By carefully selecting predictors and utilizing appropriate sorting techniques, these findings may provide valuable supplementary information to traditional diagnostic methods and psychological assessments