Advanced AI methods are rarely applied to analyze SMU’s complex effects on sleep and mental health. This study explored the associations between SMU, sleep duration, mental health and romantic relationships in youth using statistical and AI-based methods.
Method: A publicly available dataset from 705 students across 27 countries was analyzed using correlation analysis, Multiple Linear Analysis, and machine learning models including Bayesian Ridge, Random Forest, and XGBoost. Principal Component Analysis and K-means clustering were utilized for behavioral profiling, and threshold analysis was conducted to identify critical usage limits.

Significant strong negative correlations were found between daily SMU and mental health scores (r = -0.80, p < 0.001) as well as sleep duration (r = -0.81, p < 0.001). Regression analysis showed that each additional hour of SMU reduced mental health scores by 0.57 points (β = -0.5666, p < 0.001), while each additional hour of sleep increased scores by 0.19 points (β = 0.1939, p < 0.001). The model accounted for 65.6% of variance in mental health outcomes. A critical SMU threshold of 4.67 hours/ day was identified, beyond which negative effects on sleep intensified. Clustering revealed three user profiles: low-risk (32%, 3.81 hours/day), moderate-risk (41%, 4.98 hours/day), and high-risk (27%, 6.43 hours/day). Among AI models, XGBoost achieved the highest predictive accuracy (R2 = 0.954), and Bayesian Ridge showed superior generalization (CV R2 = 0.784).

SMU exceeding 4.67 hours/day marks a critical threshold for poor sleep and mental health. AI models may aid early risk detection and targeted interventions.

Hatice ULUSOY

Volkan GÖREKE