Analysis of sensitive urban form indicators of flood susceptible prediction based on machine learning models

BACKGROUND AND OBJECTIVES: Flooding is one of the biggest challenges affecting the economy and people's well-being. Previous studies have used several methods to analyze spatial data and deliver a more efficient flood response, including machine learning techniques to support decision-making in the urban planning process. However, different machine learning models serve different purposes depending on their learning processes and computation techniques. This study aims to develop a machine learning model for assessing flood risk zones to provide helpful information for city administration and planning and to support the well-being and resilience of the city's residents.METHODS: To develop a method for assessing flood risk zones and provide helpful information for city administration and planning. Eight urban factors were input into eleven multiclass classification algorithms to assess flood risk, and the results were displayed on a geographic information systems map.FINDINGS: The study discovered that the bagging decision tree algorithm model produced the best flood risk assessment model, with an accuracy of 88.58 percent compared to the government's flood simulation model results. Furthermore, rainfall, building coverage ratio, and floor area ratio were the three most important variables determining flood risk.CONCLUSION: The Bagging Decision Tree Algorithm model effectively assesses flood risk, offering valuable insights for city administration and planning. Integrating key variables such as rainfall, building coverage ratio, and floor area ratio into flood risk management strategies is crucial for mitigating the impact of floods in economically significant urban areas.