【摘要】 人工免疫识别系统(AIRS)已被证实为一种高效的分类器,并成功应用于模式识别等领域。然而AIRS存在的记忆细胞数目庞大、分类准确率低等缺陷,限制了进一步的应用。为克服这些缺陷,提出了一种基于免疫的监督式分类算法(AIUC)。AIUC首先初始化记忆细胞;然后通过对每一个训练抗原的学习,进行B细胞进化,在B细胞收敛后,优选出最佳的B细胞对记忆细胞进行更新;最后通过记忆细胞对测试数据进行kNN分类。就数据集I-ris、Ionosphere、Diabetes和Sonar分别进行的对比实验结果表明,AIUC比AIRS记忆细胞分别减小了5.6%、18%、19.6%和31%,分类准确率提高到98.2%、96.9%、78.3%和92.3%。该算法具有非线性,以及克隆选择、免疫网络和免疫记忆等生物免疫系统特征,可更好地应用于模式识别、异常检测等领域。更多还原

【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.更多还原