A compact belief rule-based classification system with evidential clustering

In this paper, a rule learning method based on the evidential C-means clustering is proposed to efficiently design a compact belief rule-based classification system. In this method, the evidential C-means algorithm is first used to obtain credal partitions of the training set. The clustering process operates in a supervised way by means of weighted product-space clustering with the goals of obtaining both good inter-cluster separability and inner-cluster pureness. Then the antecedent part of a belief rule is defined by projecting each multi-dimensional credal partition onto each feature. The consequent class and the weight of each belief rule are identified by combing those training patterns belonging to each hard credal partition within the framework of belief functions. An experiment based on several real data sets was carried out to show the effectiveness of the proposed method.