Research on Recognition and Prediction of Anxiety Problem Based on Machine Learning Method

This study utilizes machine learning methods for the identification and prediction of anxiety disorders to address a growing global mental health concern. Anxiety disorders affect millions of people globally and early detection and intervention is critical. This study utilizes a comprehensive dataset consisting of over 10,000 anonymous survey responses, observational studies, and self-reported health metrics to analyze variables such as stress levels, sleep duration, and lifestyle habits to predict the severity of anxiety disorders. The dataset was sourced from the Health Data Link service and underwent rigorous preprocessing to address inconsistencies and outliers. Four machine learning models, linear regression, random forest, XGBoost and ElasticNet, as well as the random forest-XGBoost combination, were evaluated. Performance was evaluated using <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> scores, RMSE and maximum error. The main results showed that Random Forest outperformed the other models (test <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}=0.781)$</tex>, with stress level and sleep duration being the most influential predictors. The SHAP analysis emphasized non-linear relationships, such as increased stress non-linearly exacerbating anxiety. Contributions of this study include a robust predictive framework and actionable insights for mental health monitoring. Practical applications include integration with wearable devices for real-time anxiety detection and mobile apps for personalized intervention recommendations. Future work could improve prediction accuracy by combining multi-sensors with mental health monitoring.