Predicting depressive symptoms through social support: a machine learning approach in military populations.

Perceived Social support has been consistently shown to reduce depressive symptoms among military personnel. However, limited research has explored how different types of support, emotional, informational, and instrumental, from multiple sources uniquely predict mental health outcomes. Subgroup differences based on gender, socioeconomic status (SES), and future orientation also remain under-investigated. This study used machine learning (ML) to examine the predictive effects of perceived social support on depressive symptoms among military cadets, while identifying key subgroup variations to inform tailored mental health strategies. Data were drawn from the Career Development Study of Military Personnel across four waves: SES at Wave 1 (W1), personal future orientation at Wave 2 (W2), perceived military social support at Wave 3 (W3), and depressive symptoms at Wave 4 (W4) ( = 2,978). Five ML classifiers, Random Forest, Decision Tree, Support Vector Machine (SVM), AdaBoost, and k-Nearest Neighbors, were applied to predict depressive symptoms, with model performance evaluated across full and subgroup samples. The Random Forest model achieved the highest area under the precision-recall curve (AUPRC) at 96.3% and consistently outperformed other classifiers across a range of evaluation metrics. Subgroup analyses demonstrated similarly high prediction performance, measured by AUPRC, across gender, SES, and future orientation subgroups. Feature importance analyses using the Gini index indicated that different support sources (e.g. leader, peer, senior student) played varying roles across subgroups. Machine learning approaches demonstrate high AUPRC in predicting depressive symptoms and reveal nuanced subgroup patterns in perceived social support needs. These findings can inform the development of more responsive and personalized mental health interventions in military contexts.