Rule Extraction: From Neural Architecture to Symbolic Representation
Carpenter, Gail A.
MetadataShow full item record
This paper shows how knowledge, in the form of fuzzy rules, can be derived from a supervised learning neural network called fuzzy ARTMAP. Rule extraction proceeds in two stages: pruning, that simplifies the network structure by removing excessive recognition categories and weights; and quantization of continuous learned weights, that allows the final system state to be translated into a usable set of descriptive rules. Three benchmatk 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 backpropogation network, an the C4.5 decision tree. The ARTMAP rule extraction procedure is also compared with the Knowledgetron and NoFM algorithms, that extract rules from backpropogation networks. Stimulation 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.