Rule extraction: From neural architecture to symbolic representation

Author(s): Carpenter, G.A. | Tan, A. |

Year: 1995

Citation: Connection Science, 7, 3-27

Abstract: This paper shows how knowledge, in the form of fuzzy rules, can be derived from a supervised learning neural network called furry ARTMAP. Rule extraction proceeds in two stages: pruning, which simplifies the network structure by removing excessive recognition categories and weights; and quantization of continuous learned weights, which allows the final system state to be translated into a usable set of descriptive rules. Three benchmark studies illustrate the rule extraction methods: (1) Pima Indian diabetes diagnosis, (2) mushroom classification and (3) DNA promoter recognition. Fuzzy ARTMAP and ART-EMAP are compared with the ADAP algorithm, the k nearest neighbor system, the back-propagation network and the C4.5 decision tree. The ARTMAP rule extraction procedure is also compared with the Knowledgetron and NOFM algorithms, which extract rules from back-propagation networks. Simulation results consistently indicate that ARTMAP rule extraction produces compact sets of comprehensible rules for which accuracy and complexity compare favorably to rules extracted by alternative algorithms in terms of both predictive accuracy and system complexity.

Topics: Machine Learning, Applications: Biological Classification, Medical Diagnosis, Models: ARTMAP, Modified ART,

PDF download

Cross References