Identification of key risk factors for the Polish State Fire Service with cascade step forward feature selection

Abstract

The Polish State Fire Service gathers information about incidents which require their intervention. This information is stored to document the events. However, it can be very useful for new officers training, better identification of threats and planning of more effective procedures. The identification of key risk factors for casualties among firefighters, children or other involved people was a topic of data mining competition organized as a part of 1st Complex Events and Information Modelling workshop devoted to the fire protection engineering. The task of the competition was to find ten subsets of features for ten Naive Bayes classifiers. The ensemble output was used to predict occurence of casualities. Herein, the solution description that took 5th place is presented. The proposed method used cascade step forward feature selection procedure to find features subsets.