Research and Case Study on Insurance Underwriting Decision Making Based on LSTM, SVM and Random Forest Modeling

Abstract

In recent years, the world has suffered from a variety of extreme weather events, which are gradually becoming a crisis for insurance companies and homeowners. To address the challenges of insurance companies in their underwriting decisions for a specific region, this paper proposes a binary classification model for determining whether they should be underwritten, with an output of 1 (underwritten) or 0 (not underwritten). Firstly, extreme weather events are identified by screening the number of casualties, property and crop losses and quantifying the impact of these events. Then, a system of metrics was developed, including the use of LSTM models to predict risk indicators. Correlation analyses were conducted to verify the validity of the indicator system, and then the quantitative data were fed into SVM models and random forest models to solve the classification problem. Finally, recommendations for insurance companies and homeowners were made based on the results, and two regions on different continents were selected to demonstrate the effectiveness of the model.