Building Fire Prediction Model Study Using AI

In Korea, the number of fires has been around 40,000 per year over the past decade, and is on a gradual decline. However, human and property damage, which is more important than the number of fires, is increasing due to the large scale of fire. In this study, we wanted to develop a data-based fire prediction model using artificial intelligence technology to effectively respond to the growing trend of property damage and human casualties caused by fire accidents. To this end, fire-related variables were fused on a building-by-building basis by utilizing public data being opened to the Ministry of Land, Infrastructure and Transport. Fire prediction model was developed using deep neural network model of the Multi-Layer Perceptron(MLP). The developed model showed relatively high accuracy of 87.1% as a result of the model verification through 10-fold cross validation for 60,000 random sampled units. The result of this predictive model could be used for fire prevention activities, such as management of inspection priority and inspection cycle, considering the fire risk rating of each building during safety inspection of building fires.