A methodology for fire data analysis based on pattern recognition towards the disaster management

The aim of this paper is to investigate a proposed strategy for fire disaster analysis that is implemented based on pattern recognition technique in order to achieve a methodology for disaster management. Since the fire hazard has severe effects onto human and properties, it is essential to predict and possibly prevent it. Almost every fire produces some issues, such as heat, smoke, gas, and flame, which are sensible and measurable via devices or detection systems. The fire behavior is relevant to these issues. In this research, temperature, heat radiation, and visibility (smoke) data of fire that have been obtained from Fire Dynamics Simulator (FDS) are used for analysis. The location of the fire is supposed to be in a room in the middle of the building. Also, a neighborhood containing several rooms and corridors is selected around the firing room. We have used Decision Tree learning technique to classify the data and predict the fire data. Three different scenarios have been defined to evaluate the effect of distance and time in recognition of the fire pattern. KNIME has been used for implementation of pattern recognition technique in which all fire data can be classified based on the data in the firing room. The authenticity of proposed strategy is examined via the scenarios. Statistical analysis using KNIME and MATLAB show the high accuracy (at the level of 0.9 with acceptable confidence interval) in all cases. Also, the profile of the accuracy has a logical process, according to the time and distance in most of the cases.