The purpose of this research is to develop an insurance company underwriting model by pre-processing and model construction of collected data in order to assess whether to underwrite insurance in extreme weather-prone areas and to analyse the factors that influence underwriting decisions. In this paper, data such as insurance company payout rate, number of insured, homeowners' income in selected areas, health insurance coverage, year of construction of historical landmarks, GDP per capita, and tourism income were processed, and minimum-maximum normalisation and zero-mean normalisation were used to eliminate differences in feature magnitude. In the model construction, the logistic regression model and BP neural network were used to derive the underwriting probability, and the three main factors of natural disaster risk assessment, payout rate, and homeowner influence were introduced, and the relationship between these factors and the underwriting probability was described by the logistic distribution function, which determined that the underwriting probability should be underwritten when the underwriting probability Pi≥0.5. The natural disaster risk assessment model selects disaster factors, disaster environment and disaster receptors as assessment indicators, constructs the CEDP model, and calculates the weights using the AHP model, the Topsis model and the CPITIC weighting method, and carries out fuzzy clustering analysis on the CEDP index of 25 regions to classify the risk into five levels. For insurer profits, solvency, premium income and claims ratios were analysed, ratios based on annual claims to total premiums were calculated, and the impact of homeowners' incomes, property and historic buildings on underwriting decisions was assessed. Insurance company underwriting decision factors in areas prone to extreme weather are ultimately determined.