Exploring the relationship between social media use, eating behaviors, and depression in young women: a machine...

The increasing prevalence of social media use among young adults has raised concerns regarding its potential associations with mental health and eating behaviors. The aim of this study was to investigate the associations between social media use, eating behaviors, and depressive symptoms among young women and to evaluate the performance of machine learning algorithms in modeling these cross-sectional associations compared with conventional regression approaches.

A cross-sectional study was conducted among young women aged 18-34 years (n = 4,692) using a multi-site, face-to-face convenience sampling strategy. The survey collected data on BMI, patterns of social media use, the Beck Depression Inventory (BDI), and the Scale of Effects of Social Media on Eating Behavior (SESMEB). In addition to standard statistical analyses, various ML models, including Random Forest and Extreme Gradient Boosting (XGBoost), were applied to model SESMEB and BDI scores. Model performance was evaluated using internal validation procedures, including a train-test split and cross-validation, and assessed using root mean square error (RMSE) and mean absolute error (MAE).

The mean age of participants was 23.8 years, and the mean BMI was 22.3 kg/m². Significant differences in BMI, SESMEB, and BDI scores were observed across social media platform use groups. Users of Facebook, TikTok, and Snapchat reported significantly lower SESMEB scores (p < 0.001). ML models identified BMI, snack intake, and social media use as influential variables associated with SESMEB and BDI scores. Random Forest showed the best modeling performance in the test set for both SESMEB and BDI scores, with XGBoost performing comparably.

Social media use was associated with eating-related behaviors and depressive symptoms in young women. Within the study sample, ML models demonstrated modest modeling performance for SESMEB and BDI scores and may provide a complementary framework for examining complex associations among behavioral and psychological variables. However, these findings were based on internal validation only and require confirmation in independent external samples. Further research is warranted to clarify the direction of these associations, explore platform-specific differences, and evaluate the external validity of these models.