Fuzzy Inference with Sequential Fuzzy Indexed Search Trees

In this paper, a new method is presented for fuzzy inference-based classification. Its base idea lies in the dimensional decomposition of the problem space: the proposed method builds a structure from the training data in which the search among the fuzzy rules is done dimension by dimension, and thus, the number of rules that are needed to be evaluated is gradually restricted. The structure has a layered architecture, where each layer corresponds to a given dimension of the input data and contains a set of fuzzy membership functions, each with a self-balancing binary search tree to quickly identify the relevant fuzzy sets. These are implemented using indexing arrays to enhance the operating speed.