Development and Verification of a Fire Occurrence Prediction Model with Social-Architectural Factors

Abstract

In this study, we aim to develop a fire occurrence prediction model at the administrative district level by incorporating social-architectural factors. Based on data on social and architectural factors and the number of fire occurrences from 2015~2021 in various cities, counties, and districts, multiple machine learning algorithms (multilayer perceptron, LASSO regression, and random forest) were employed to implement and compare the performance of the fire occurrence prediction models. The results indicated that the model utilizing the random forest algorithm exhibited the best prediction performance. Furthermore, performance validation using 2022 data showed that out of 247 administrative districts, 171 had an error rate of 20% or less.