A Proposed Ensemble Model for The Prediction of Coronavirus Anxiety Scale of Migrant Workers

Abstract

This study aimed to evaluate the potential negative effects of the scattered migrant worker population on the anxiety level by estimating the coronavirus anxiety scale (CAS) of the COVID-19 anxiety scale with Gradient Boosting Tree (GBT). In this study, a public data set achieved from a questionnaire [developed using the Coronavirus Anxiety Scale (CAS)] was used to conduct on 1350 people over phone calls. GBT model was constructed for predicting the CAS score of migrant workers based on input variables including demographical data. Hyperparameters of the GBT model were tuned using Optimize Parameters (Evolutionary) operator, which seeks the optimum values of the selected parameters by an evolutionary computation approach. Hyperparameters of the GBT model were 50 for the number of trees, 5 for minimal depth, 0.044 for learning rate, and 1.0E-5 for minimum split improvement. A total of 1500 people, 758 (56.1%) male, and 592 (43.9%) female, participated in this study. The experimental findings demonstrated that the GBT yielded a root mean square error of 3.547±0.235, the absolute error of 2.943±0.154, relative error lenient of 31.54%±0.82%, squared error of 12.623±1.691 and correlation of 0.577±0.130. Variable importance values for each input were calculated from the model-based results of the GBT model. The largest importance was achieved for income and the lowest was estimated for Covid-19 Infection. The calculated importances can be evaluated the potential impacts on the CAS score. In future works, different algorithms can be built for detailed predictions about COVID-19-related anxiety levels