Flood Probability Prediction based on Regression Algorithm and Integrated Learning

Flood is a natural disaster that seriously threatens human life and property and ecological environment. In this paper, we construct a flood probability prediction model based on regression algorithm, feature degradation and integrated learning. The data covers monsoon intensity, topographic drainage and other indicators, more than 1 million. After preprocessing, transformation and normalisation, Pearson correlation analysis is used to identify factors significantly affecting flood risk, such as infrastructure condition, topographic drainage capacity, etc., and preventive measures are proposed. As for the model, the GBDT model has limited effect and is replaced with TruncatedSVD dimensionality reduction combined with logistic regression and L-BFGS algorithms with an accuracy of 77%. The flood probability prediction model compares linear regression, ridge regression and MLP, and linear regression performs the best, with an R² value of 0.73563. The results of this paper provide a new method for flood risk assessment, and deep learning models such as LSTM can be explored in the future in combination with GIS technology.