Wu Qingfeng

Research on adaptive distributed intrusion detection system model based on Multi-Agent

To overcome the interdependence between the traditional intrusion detection system components, high false-alarm rate of the duplicated ones from the same invasion, and the inability of adaptive replacement of algorithm for mining when testing environment changed, this paper puts forward an adaptive distributed intrusion detection system model based on Multi-Agent, whose mining algorithm module employs Joint Detection mechanism, while dynamic election algorithm for the recovery mechanism, which improves the system adaptive ability to the external changes.

The research of missing value estimation of gene sequence based on improved KNN

Gene based data mining has been received wider and wider attention as gene carries genetic information of living creature. While mining gene information, one of the tasks is to estimate the missing values reasonably and effectively, so as to reflect the original information of gene sequence. By analyzing the theory of KNN (K nearest neighbor algorithm), an improved KNN for gene sequence was proposed, which resolves the problem of missing values while mining gene data. Results show the feasibility of the algorithm with experiments using data from genbank.

A Survey of Application-Level Protocol Identification Based on Machine Learning

Application-level protocol identification has attracted great interests in academia and become a relatively independent research realm. With the rapid development of Internet and the protocols complicated day by day, the traditional port-based application-level protocol identification algorithms become inaccurate. Machine learning is a hot research in the hour. Many researchers have taken the method into consideration in application protocol recognition. The high correct rate and wide applicability make it promising. In the paper, some basic conceptions of protocol identification are introduced and some important algorithms of machine learning used in application-level protocol identification are summarized in three main way. The applicability is summarized by comparison. In the end, some disadvantages in the method and future research directions are posed.

SCAD algorithm and its application in analyse of bank loan

Feature weighting can be considered an extension of feature selection. Traditional methods of feature weighting assume that feature relevance is invariant over the tasks domain. As a result, they learn a single set of weight for the entire data set. In this paper, a proposed algorithm has been used, which is called Simultaneous Clustering and Attribute Discrimination (SCAD) and performs clustering and feature weighting simultaneously. First, the algorithm is analyzed in detail, on this basis, through a series of compare experiments, confirms this algorithm to have the high clustering precision; Finally, the algorithm is applied in the analyzing of the bank loan repaid information, that can efficiently discover the weight association of the main factors in loan information and realize potential customer.