Skip to Main Content
In this paper we present a neural-fuzzy approach to rule extraction, which is based on a generic definition of incremental perceptron and a new competitive learning algorithm we recently developed. It extracts a suitable number of rule patches and their positions and shapes in the input space. Initially the rule base consists of only a single fuzzy rule; during the iterative learning process the rule base expands according to a supervised spawning-validity measure. The rule induction process terminates when a stop criterion is satisfied. The proposed approach will be effective in dynamic data-mining applications. To demonstrate the effectiveness and applicability of our algorithm, we present a simulation result. This algorithm is currently being tested on a number of data sets from biology and the Web.