Decision rule learning from stream of measurements—a case study in methane hazard forecasting in coal mines

Abstract

The approach based on the Very Fast Decision Rules algorithm in application to prediction of alarm state resulting from methane hazard in coal mines is presented in this work. The approach introduces the modification of rule induction process due to application of the Correlation rule quality measure. An evaluation of the introduced method on a real life stream data collected from coal mine sensors is performed. The results show advantages of the introduced method considering both the classification quality and the rule-based knowledge representation.