Research On Feature Selection Technologies Of Artificial Immune NETwork In Streaming Data Environment

Artificial Immune Network is a bio-inspired computational model that uses ideas and concepts from the immune system theory to solve engineering and scientific problems. Nowadays, it has been an important research field of the theory and application of artificial immune systems. This dissertation focuses on artifical immune network and its technology which is used for feature selection in streaming data environment, and makes the following three main contributions:1. In large-scale, dynamic, time-varying streaming data environment, like web electronic commerce, medical surveillance, sensors, financial monitoring etc, the computational cost of present immune network model will increase so greatly that it can not meet the real-time needs. To solve the problem, a novel artificial immune network model named IFSaiNET which is used for feature selection in streaming data is advanced. It uses an outlined data set —immune memory antibody set which is much smaller than the size of streaming data set to reduce streaming data. The set can help IFSaiNET obtain overall variational characteristic of the streaming data. In addition, the network model provides incremental strategy by "window" mechanism, and has the ability of allowing dynamic tracking of ever-increasing large scale information. The results of experimental evaluation show that the new approach has significant superiority on running time.2. In fact, IFSaiNET uses a little-scale immune memory antibody set to reflect antigen data collection, so that it can select data feature. Since the mechanism of immune clone algorithm doesn't have a method to evaluate the performance of the artificial immune network model through the network's topology information feature, this dissertion proposes a topology structure analysis technique of artificial immune network which is based on complex network. Firstly, it forms the network structure of antibody and antigen. Secondly, it chooses the community structure of complex network model to be the benchmark to evaluate the study performance of the artificial immune network model. Thirdly, it advances a method to evaluate the feature selection performance of the artificial immune network model, through comparing the antigen network before data selection and the community structure of the artificial network model after data selection. Also, it proves the artificial immune network model can maintain stability of the network topology structure before and after feature selection. At the same time, to analyze the influence on studing performance of IFSaiNET from the imbalancing problem of dataset, a method to evaluate the imbalancing degree based on the community structure is proposed. It gives two key elements named coupling and cohesion which determine imbalancing degree of the dataset. The result shows the effect of evaluating the imbalancing degree of the dataset using community structure is prominent, and...