Explainable Machine Learning for Predicting Depression Risk Using Sleep Patterns, Self-Esteem, and Psychosocial Factors

Depression presents a substantial public health challenge, shaped by interacting behavioural, psychological and social conditions.Earlier identification of vulnerability requires approaches that capture these relationships while remaining interpretable and appropriate for the populations assessed.This proposed study will examine whether sleep patterns, self-esteem and psychosocial factors improve depression prediction beyond demographic and health-related characteristics.Validated measures will characterise sleep quality, duration and regularity, self-esteem, perceived stress, loneliness and social support.An interpretable regression baseline will be compared with selected machine learning algorithms using domain-specific and combined predictor sets.Model development will incorporate missing-data procedures, leakage prevention, tuning and independent evaluation.Discrimination, calibration, prediction error and subgroup performance will assess reliability, while complementary explanation methods will examine influential features, nonlinear patterns and explanation stability.Analyses will address overlap between sleep predictors and depression questionnaire items.Longitudinal data will support future-risk prediction; concurrent measurements will support identification of elevated symptoms only.Findings will clarify the incremental value and limitations of combining predictor domains, without treating explanations as causal evidence.External validation and evaluation of clinical usefulness will remain necessary before routine implementation.