Designing Distributed Chi-Fuzzy Rule based Classification System

Fuzzy Rule based Classification Systems (FRBCSs) are an important area of fuzzy logic and fuzzy sets. FRBCSs provides interpretable models with good classification rate. In the presence of large number of instances with high dimension in training data, the classical Chi’s FRBCSs’ rulebase becomes huge causing the classical model to be computationally expensive which also leads to the degraded interpretability. In this work, we propose the ‘Distributed Chi Fuzzy Rule Based Classification Systems’ (DCHI-FRBCSs). The proposed approach distributes the n-dimensional data into n-nodes where each node forms its own 1-dimensional rules for the dimension. The output from each node is the rule weights which are then aggregated to form the final rule weights, the class with highest final rule weight is the final output of the model. Aggregators play an important role in combining the information from several sources. Andness-directed OWA selects desired level of andness between attributes for OWA aggregator. In this paper, we explore an extension of DCHI-FRBCSs called ‘Andess-directed Distributed Chi Fuzzy Rule Based Classification System’ (α-DCHI-FRBCS) which incorporates andness-directed OWA operator as the aggregator function. Results over UCI datasets verify the superiority of our methodology.