Rule Extraction from Artificial Neural Network with Optimized Activation Functions

摘要：

A novel method of rule extraction from artificialneural network with optimized activation function isproposed.Weight-decay approach is used in trainingand the unnecessary connections in the neural networkare pruned at the cost of an increase in the errorfunction within a predetermined limit.A penalty term isadded in the activation function to facilitate the valuesof hidden and output nodes to have betterapproximation to 0 or 1,which is of great help insymbolic rule ex.traction in neural network With theoptimized activation function,the rule extractionbecomes much easier and simpler.Rule extraction hasbeen experimented on two public datasets of Iris andBreast-cancer,which results showed that the proposedmethod has a better rule overcast accuracy than thecommonly used methods,such as Decision TreeAlgorithm C4.5 and RX algorithm.

关键词： rule extraction artificial neural network optimized activation function

作者: WANG Jian-guo YANG Jian-hong ZHANG Wen-xing XU Jin-wu

作者单位: Mechanical Engineering School,University of Science and Technology Beijing,Beijing China 100083;Mech Mechanical Engineering School,University of Science and Technology Beijing,Beijing China 100083 Mechanical Engineering School,University of Science and Technology Inner Mongolia,Baotou China 01401

会议类型: 国际会议

会议名称: 2008 3rd International Conference on Intelligent System and Knowledge Engineering(第三届智能系统与知识工程国际会议)(ISKE 2008)

会议地点: 厦门

会议语种:英文

页码: 873-879

在线出版日期: 2008-11-17（万方平台首次上网日期，不代表论文的发表时间）