Wang Ziqiang,Feng Boqin
(School of Electronics and Information Engineering,Xi'an Jiaotong University,Xi'an 710049,China)
Abstract:To efficiently mine the classification rule from databases,a novel classification algorithm based on immune algorithm was proposed.The core of the immune classification algorithm is as follows.The rule antecedent is encoded as fixedlength chromosome;The fitness function is calculated according to minor misclassification ratio,simplicity and consistency of rules, and coverage ratio of training examples;A vaccination is accomplished by modifying genes on some bits in accordance with minimal fitness function which serves as prior knowledge;Immune selection is accomplished by testing whether a serious degeneration has happened in the evolutionary process and annealing selection.Meanwhile, a rule pruning procedure based on information gain was designed for improving the comprehensibility of classification rule mined.The algorithm has been compared with RISE and OCEC algorithms with five benchmark datasets from UCI data set repository. Experimental results show that the proposed algorithm not only has faster convergence speed, but also can achieve higher prediction accuracy with less number of rules.
Keywords:data mining;classification rule;immune algorithm;information gain