(15c) Regression Analysis and Machine Learning for the Prediction of Fire Property Damage Based on Fire Type, Weather Conditions and Site Response Characteristics

In modern society, not only the promotion of urbanization and industrialization as the population increases, but also the factors that influence the fire outbreak are diversified due to the development of science and technology, showing a complex relationship. In cities with high population due to urbanization and industrialization, fire occurrence characteristics according to the use of buildings such as residential facilities, industrial facilities, multi-use facilities, building structure, and building size, as well as technological fire factors such as mechanical, electrical, and gas leakage. Fire has been steadily occurring, and an average of about 42,000 fires have occurred annually over the past decade, and property damage and human casualties have also been steadily increasing. Fire statistics yearbooks are published every year, but as fires occur through complex relationships between various factors, the need for fire research on new concepts using the 4th industrial revolution technology to analyze and predict fire occurrence factors beyond conventional statistics is increasing.

This study aimed to predict the size of fire damage by regression analysis of the characteristics of fire factors, and to continuously reduce damage to property and human life by preparing preventive measures based on a quantitative model. Data were collected from the National Fire Information System (NFDS), and variables affecting property damage were selected as fire site arrival time, fire site suppression time, ignition heat source, ignition factor, location, weather, and building structure. Prior to machine learning-based predictive modeling, SPSS, a statistical analysis program, was used to analyze the correlation between variables. In particular, linear regression analysis and nonlinear regression analysis were conducted to identify the increase in property damage by creating a prediction model according to fire type and extinguishing time, and to complete an improved prediction model by using and learning the analyzed statistical data in machine learning-based prediction model.