# Identifying influencing factors associated with sleep quality in undergraduates based on partial least squares regression and XGBoost

## TL;DR

This study identifies factors affecting sleep quality in Chinese undergraduates using statistical models and finds mental health, smartphone use, and dormitory noise to be key influences.

## Contribution

The study introduces an integrated PLSR-XGBoost framework to handle multicollinearity and capture nonlinear relationships in sleep quality factors.

## Key findings

- Mental health status was the most significant factor affecting sleep quality.
- Smartphone dependence and noisy dormitory environments were strongly linked to poor sleep.
- The XGBoost model showed good performance in classifying sleep quality (AUC = 0.818).

## Abstract

This study aimed to identify the influencing factors associated with sleep quality among undergraduates in Jiangsu, China, and to explore their complex relationships.

A cross-sectional survey was conducted online between October and November 2022. A total of 7,062 valid participants (aged 20.1 ± 1.3 years) were included, and a complete case analysis was performed. Sleep quality was assessed using the Pittsburgh Sleep Quality Index (PSQI), with a score exceeding 7 denoting poor sleep quality. To overcome the limitations of traditional statistical methods in handling multicollinearity and capturing complex, nonlinear associations, Partial Least Squares Regression (PLSR) was used to quantify linear relationships between influencing factors and quantitative PSQI scores, while the eXtreme Gradient Boosting (XGBoost) model and SHapley Additive exPlanations (SHAP) analysis were utilized to identify nonlinear influencing factors and their interactions associated with binary classification of sleep quality.

The prevalence of poor sleep quality was 26.5% (95%CI: 25.5 to 27.6%). Mental health status was the most influencing factor, and those with good mental health had better sleep quality (β = −0.853, 95%CI: −1.079 to −0.627, p < 0.001). Smartphone dependence (MPAI: β = 0.043, 95%CI: 0.038 to 0.048, p < 0.001) and a noisy dormitory environment (β = 0.627, 95%CI: 0.420 to 0.835, p < 0.001) were significantly associated with poorer sleep quality, whereas psychological resilience (CD-RISC: β = −0.017, 95%CI: −0.023 to −0.012, p < 0.001) emerged as a negatively associated factor. Other significant associations were found for smoking (β = 0.496, 95%CI: 0.276 to 0.716, p = 0.002), drinking (β = 0.594, 95%CI: 0.420 to 0.768, p < 0.001), and harmonious interpersonal relationships (β = −0.172, 95%CI: −0.251 to −0.094, p = 0.002). The XGBoost model showed good discriminative performance (AUC = 0.818, 95%CI: 0.795 to 0.841). SHAP analysis revealed nonlinear patterns, such as a U-shaped relationship between BMI and sleep quality.

The integrated PLSR-XGBoost framework effectively handled multicollinearity without discarding variables and provided a more comprehensive understanding from both linear and nonlinear perspectives. The findings support the development of comprehensive and tailored interventions targeting specific influencing factors, such as mental health support, dormitory noise management, smartphone use modification, and resilience-building programs, offering an empirical foundation for sleep health promotion initiatives in university settings.