Feature selection for situation recognition in Fuzzy SOM-based Case-based Reasoning

Abstract

Case-based Reasoning (CBR) is a problem solving approach applied to different cognitive systems for planning, decision making, etc. This approach benefits/utilizes the solutions of previous similar problems for solving a new problem. Situation recognition as an important process in CBR provides knowledge about actual problem including the situation of system. The system's situation defined with a set of characteristics/features, models the scene and illustrates an internal structure of the system. The situations are learned as experiences by the system for further usage. Dealing with a large amount of experiences as well as imprecise, uncertain, and redundant data (characteristics) is a challenge for situation recognition. Investigation of all characteristics of a situation for defining the actual problem may decrease the system performance in terms of recognition accuracy and computational complexity. Therefore, using an appropriate method to discard irrelevant characteristics may improve situation recognition approaches. Here, an improved CBR based on Situation-Operator Modeling (SOM) and Fuzzy Logic (FL) is applied as the base CBR. The fuzzy SOM-based CBR benefits an effective knowledge representation approach to support different situation recognition levels and handles uncertainties. This contribution aims to address the effects of feature selection in dealing with data redundancy in fuzzy SOM-based CBR. A feature selection approach based on Rough Set Theory is then applied to the CBR to find an optimal set of relevant characteristics for the situations. Finally, the proposed CBR approach is realized using an experimental application (driving maneuvers) to show the effectiveness of the feature selection on situation recognition.