Improving Of Artificial Imune Classification And Anomaly Detection Algorithms

The immune system is one of the main systems of living systems. The self and non-self non-linear adaptive network constructed from a rich variety structures of antibodies, playing an important role in dealing with complex and changing environment. Inspired by the principles of the immune system, artificial immune algorithm developed and provides a good evolutionary learning mechanism for scholars, because it has some specialties of diversity, tolerance, distributed parallel processing, self-organization, self-learning, adaptive and robustness, it's able to express the learning of knowledge clearly, and has memory of content function. The function of natural defense system it mimics is strikingly similar with the anomaly detection system's function of distinguishing the normal and abnormal, and itself has well learning and memory ability and strong robustness. In short, artificial immune algorithm provides a new choice to solve the problems such as anomaly detection and classifier structuring.Firstly, in this paper, a number basic concepts, characteristics and mechanism of biological immune system is described, some classic algorithms of this field and its merits and demerits are introduced. Secondly, aims at the problem of "empty" in the traditional negative selection algorithm, this paper improves the demerit of antibody unity by simulating the characteristics of antibody diversity in the immune algorithm, and advances an improved immune anomaly detection algorithm with variable-size detector, whose validity were verified through experimental analysis. Finally, on the problem of low accuracy under a small amount of training data, this paper imports the learning mechanism of semi-supervised and the thinking of the voting decision, then, a semi-supervised classification immune algorithm was raised and verified experimentally.