用于异常检测的免疫实值检测器优化生成算法

Optimization algorithm for immune real-value detector generation

针对已有实值可变半径检测器生成算法的不足,提出一种优化的检测器生成算法。通过对检测器生成过程的统计分析,给出了基于假设检验的检测器生成过程,并将假设检验的结果作为算法结束的一个控制参数,有效减少了冗余检测器的产生。同时,算法充分利用自体空间的分布,优化检测器生成的中心位置,扩大检测器的半径,尽可能生成覆盖范围大的检测器,提高检测性能。通过人工合成数据集2DSyntheticData以及实际的Iris数据集和Biomedical数据集对算法进行了验证。实验结果表明,本算法用于异常数据检测,提高了检测率,所需的检测器数量减少,整体检测性能较优。

A new optimized detector generation algorithm is proposed to overcome the shortcomings of available real-value variable-radius detector generation algorithms.By statistic analysis of the detector generation,a hypothesis testing based detector generation process is proposed.The result of the hypothesis testing is taken as one of the control parameters to end the algorithm,thus,it can effectively avoid the generation of redundant detectors.Meanwhile,the algorithm makes full use of the distribution of self-space,optimizes the center position and expands the radius of the detectors in order to generate the detector with large coverage.The 2DSyntheticData,the actual Irish data set and biomedical data set are used to test the algorithm.Experiment results show that the algorithm performs very well that it improves the detection rate,reduces the number of required detectors.