Research on Predicting Ignition Factors Through Big Data Analysis of Fire Data

Abstract

The National Fire Agency collects data such as reporting, dispatch, and suppression of fire incidents in the country every year. However, it remains at the basic level of analysis of frequency or cause of fire that occurs. In response, about 46,000 fire data throughout the country were analyzed in 2018 to predict the cause of ignition in the event of a fire. The majority of data recorded in a single fire event is unsuitable for analyzing a fire, and there is no value or data irrelevant to the cause of the fire. Thus, the data was refined to about 30,000 cases, excluding meaningless values. In addition, because it is a study to infer the unknowns of the ignition factors, data that are directly related to the ignition factors were excluded. As a result, an artificial neural network algorithm was applied to infer the ignition factors using about 10 data per fire accident, and the prediction accuracy was about 80%. Rather than determining the ignition factors through this data analysis, it is expected to provide information to fire extinguishing groups and help to increase the level of fire extinguishing. In addition, in order to use the collected data for analysis and prediction, the structure of the database needs to be improved. If the system is improved with artificial intelligence in fire inspection, it is expected that the analysis results of artificial intelligence will be provided in real time by inputting information at the site.