Background: Sleep quality plays a crucial role in adolescents' health and well-being. The aims of this study is to predict the sleep quality of female students in Iraq by using interpretable machine learning algorithms.
Methods: A cross sectional data-driven study was carried out on 890 female students in Iraq in 2026. After data cleaning, variable preprocessing, and class balancing using the Synthetic Minority Oversampling Technique, three machine learning models were trained and tested for predicting sleep quality: Logistic Regression, Random Forest, and XGBoost. Model performance was evaluated by determining Accuracy, F1 score and Area Under the Receiver Operating Characteristic Curve (ROC-AUC). SHAP was used to explain and assess the contribution of the most influential factors associated with sleep quality, to boost the interpretability of the best-performing model.
Results: The machine learning models showed good performance in sleep quality prediction. In terms of the classification performance, among the evaluated models, XGBoost was the highest with accuracy 83.1% and macro F1-score 81.6%, and Random Forest was the highest with ROC–AUC 0.904 in terms of discriminative ability. SHAP analysis showed sleep duration, age, physical activity status, psychological domain of quality of life and mobile phone use as the most powerful predictors of sleep quality.
Conclusion: the results of this study confirm that the use of machine learning models in conjunction with interpretable techniques of artificial intelligence like SHAP has improved the predictive performance and also helped to establish the most important determinants of sleep quality. This strategy could help shape the creation of intelligent screening systems in the schools and tailored prevention interventions that will enhance the sleep health of adolescents.