An Immune Based Supervised Classifier
PENG Ling-xi,LIU Xiao-jie,LI Tao,LU Zheng-tian,ZENG Jin-quan,LIU Cai-ming
2008-01-01
Abstract:Artificial immune recognition system(AIRS) had been proved a highly effective classifier,and successfully applied to pattern recognition.However,the huge size of evolved memory cells pool and low classification accuracy limited the further applications of AIRS.In order to overcome these limitations,a supervised artificial immune classifier,referred to as AIUC,was presented.The implementation of AIUC included: initially,a pool of memory cells were created.Then,through the learning of each training antigen,B-cell population was evolved until the B-cell population was convergent,and the memory cells pool was updated by the optimal B-cell.Finally,classification was accomplished by majority vote of the k nearest memory cells.Compared with AIRS,AIUC showed the improvements for the percentages reduction of memory cells pool by 5.6%,18%,19.6% and 31%,respectively,meanwhile,the classification accuracies increased to 98.2%,96.9%,78.3%,and 92.3%,for the famous Iris dataset,the Ionosphere dataset,the Diabetes dataset,and the Sonar dataset,which were used for testing classification algorithm,respectively.In addition to its nonlinear classification properties,AIUC possessed biological immune system properties such as clonal selection,immune network,and immune memory,which could be better used to pattern recognition,anomaly detection.