Generalized analytic rule extraction for feedforward neural networks

This is paper suggests the "Input-Network-Training-Output-Extraction-Knowle dge" framework to classify existing rule extraction algorithms for feedforw ard neural networks. Based on the suggested framework, we identify the majo r practices of existing algorithms as relying on the technique of generate and test, which leads to exponential complexity, relying on specialized net work structure and training algorithms, which leads to limited applications and reliance on the interpretation of hidden nodes, which leads to prolife ration of classification rules and their incomprehensibility. In order to g eneralize the applicability of rule extraction, we propose the rule extract ion algorithm Generalized Analytic Rule Extraction (GLARE), and demonstrate its efficacy by comparing it with neural networks per se and the popular r ule extraction program for decision trees, C4.5.