Research on Sustainability of Property Insurance Based on LSTM Model

Abstract

Losses from extreme weather events are increasing in recent years, and climate change will drive premiums higher. At present, the adjustment of insurance policies by insurance companies, as well as the protection of real estate siting and ancient buildings, are the problems that we need to solve now. To explore the circumstances under which insurance companies undergo insurance, this paper identified two severely affected areas and collected monthly insurance claims data from 2010 to 2023. Based on the collected data, an LSTM is established to derive the amount of compensation, and then determine whether to underwrite. The research centers on "how and where to choose a site and whether to build in a certain place", we determined 5 aspects of research in natural conditions, transportation conditions, infrastructure, surrounding environment and economic factors, and selected 17 secondary indicators; firstly, the paper quantified the secondary indicators into 3 parts of proximity, remoteness and Neutral factor 3 parts. Firstly, using EDA and feature engineering & insurance factor quantitative measurement to find out the required insurance factor F. Then using Kmeans cluster analysis to classify the 19 regions, and using Random Forest to find out the weights of the 17 indicators. Secondly, using TOPSIS comprehensive evaluation model to find out the scores of the 19 regions, and finally determining the highest-scoring the "leader" with the highest score is finally determined as the optimal region, For the research of developing building protection model, this paper identifies four aspects of research from basic information, economic contribution, historical contribution and architectural information. Finally, the Word to vector word embedding vector model is used to number the words after the text is transformed; then the LSTM is used to do the text binary classification to get the emotional percentage of each area, and the emotional percentage is used as a new column of indicators; the PCA is used to extract the variance contribution rate of each indicator; then the MinMax variance contribution rate is used to carry out the normalization and the sigmoid function for weighting; finally, use AHP to find out the score, and from the results, it can be seen that the regional conservation model of Yellowstone National Park is the most reasonable.