不完备数据下的免疫分类算法

摘要

Artificial Immune Recognition System (AIRS) that is inspired by natural immune system has been developed as an efficient classifier. But the number of memory cells is too large and the accuracy of AIRS is not high, especially in the case of incomplete data. To solve the problem, this paper presents Immune Classification Algorithm Under (ICAU) incomplete data. It introduces semi-supervised learning, classifier fusion and vote to decision ideas, uses multiple AIRS classifiers to learn to refine each other. In the UCI data sets, experimental results prove the validity of the ICAU algorithm.%人工免疫识别系统(AIRS)是受生物免疫系统的启示而研发的一种比较有效的分类器,但也存在记忆细胞数目过于庞大,分类精度不高,特别是在数据不完备的情况下,分类精度低等缺陷.为了解决这个问题,提出了一种不完备数据下的免疫分类算法(ICAU),算法引入半监督学习机制和分类器融合投票决策的思想,利用多个AIRS分类器互相帮助学习训练,来提高AIRS在不完备数据下的分类精度.在UCI数据集上进行了实验,结果验证了ICAU算法的有效性.