A self-organizing feature map-driven approach to fuzzy approximate reasoning

In literature, there has been an increasing interest in the fuzzy approximate reasoning (FAR) via fusion of neural networks and fuzzy logic under the name of fuzzy neural network (FNN). Therefore, FNN usually provides a main theoretical basis for FAR. Most of the existing FNN models have been proposed to implement different types of single-staged fuzzy reasoning mechanisms. The single-staged FAR, however, is far short of effectively solving complicated decision-making problems. Rather, we need a multi-staged FAR in which the consequence of a rule in one reasoning stage is passed to the next stage as a fact, leading to building up a high level of intelligence to solve problems. In this sense, we propose a new multi-staged FAR named SOFAR (self-organizing FAR) by integrating self-organizing feature map (SOFM) and fuzzy logic. From the stipulated input–output data pairs, the proposed SOFAR can generate appropriate fuzzy rules via SOFM and modified back-propagation driven parameter modifications. To illustrate the performance of the proposed SOFAR, we first used a simulated data from Takagi and Hayashi [Takagi, H., & Hayashi, I. (1991). NN-driven fuzzy reasoning. International Journal of Approximate Reasoning, 5, 191–212]. Then a real data set was adopted from a construction of retaining wall in urban area, applying the proposed SOFAR to obtain promising results in terms of error rate and statistical tests.