Using Data Mining Techniques to Assess the Impact of COVID-19 on the Auto Insurance Industry in China

Since coronavirus disease 2019 (COVID-19) was discovered at the end of 2019, the whole world has been severely affected. The insurance industry, regarded as an important factor in recovery, has also been affected by COVID-19. However, effective data mining techniques have rarely been utilized in the insurance industry in China, especially under the circumstances of COVID-19. Although some traditional statistical analysis methods have been applied to this area, the limitation of the lack of data distribution still cannot be efficiently overcome. With the machine learning technique proposed in this thesis, this limitation can be solved by using a stacking model with great generalization ability. In this research, the ElasticNet, LightGBM, and Random Forest approaches were employed as base learners; ridge and LASSO regression were used as meta-models to increase the prediction accuracy; and the SHAP value was utilized to explain the impact of COVID-19 on the insurance industry in China. The stacking meta-model in this thesis has a mean absolute percentage error (MAPE) of 12.57134, whereas the average value in the past week is 21.50972, and the MAPE of ElasticNet is 22.57935. In conclusion, COVID-19 affects the auto insurance industry in China.