Predicting depression and anxiety of Chinese population during COVID-19 in psychological evaluation data by XGBoost.

ABSTRACT

BACKGROUND: Due to the onset of sudden stress, COVID-19 has greatly impacted the incidence of depression and anxiety. However, challenges still exist in identifying high-risk groups for depression and anxiety during COVID-19. Studies have identified how resilience and social support can be employed as effective predictors of depression and anxiety. This study aims to select the best combination of variables from measures of resilience, social support, and alexithymia for predicting depression and anxiety. METHODS: The eXtreme Gradient Boosting (XGBoost1) model was applied to a dataset including data on 29,841 participants that was collected during the COVID-19 pandemic. Discriminant analyses on groups of participants with depression (DE2), anxiety (AN3), comorbid depression and anxiety (DA4), and healthy controls (HC5), were performed. All variables were selected according to their importance for classification. Further, analyses were performed with selected features to determine the best variable combination. RESULTS: The mean accuracies achieved by three classification tasks, DE vs HC, AN vs HC, and DA vs HC, were 0.78, 0.77, and 0.89. Further, the combination of 19 selected features almost exhibited the same performance as all 56 variables (accuracies = 0.75, 0.75, and 0.86). CONCLUSIONS: Resilience, social support, and some demographic data can accurately distinguish DE, AN, and DA from HC. The results can be used to inform screening practices for depression and anxiety. Additionally, the model performance of a limited scale including only 19 features indicates that using a simplified scale is feasible.