Classification Rules Discovery Based On Local Search And The Application In Intrusion Detection

Data Mining is a domain that tries to extract knowledge and interesting information from very large-scale databases. This knowledge is hidden, unknown, but potentially useful. Classification rules discovery is a procedure to construct a classifier through studying the training dataset. It is a very important part of Data Mining and Knowledge Discovery. In essence, the goal of classification rules discovery is to discover knowledge that is highly accurate, comprehensible and interesting. This paper has done some research on classification rules discovery based on Local Search. We also do a further research on classification rules discovery based on Local Search and Ant algorithm combined together. Local Search is a common approximate search algorithm. In this paper we study the key technology to the implementation of multi-start Local Search in classification rule discovery, including confirm the choice tactics of the starting point, the expression of answer, definition of the neighborhood, design of goal function and the tactics of local search, ect. On this basis, we explained the basic thought of classification rules discovery based on multi-start local search method. We have carried out the experiment of the algorithm for four public data sets, and evaluate the algorithm's performance. The results show the algorithm can obtain classification rules with high predictive accuracy, and the rules discovered by the algorithm are considerably simple and easy for user to understand.The local search method for classification rules discovery can obtain good classification rules with high predictive accuracy, however the running time of the algorithm is relatively long. On the basis of analyzing the pluses and minuses of the classification rules discovery method based on local search, we introduced basic thought of Ant algorithm, namely pheromone and it's updated, in the proceed of local search, let pheromone guide the path of search. On the basis of this thought, we have put forward and found the method for classification rules discovery based on local search and Ant algorithm combined together. The experiment indicates this method can obtain better trading off between rule quality and running time of algorithm. At present constant enlargement in network size and hacker attack means complicated day-by-day, the demands of people for the network security grow with each passing day. Intrusion detection technique as an important branch of the network security is paid close attention to by more and more people too. The application of data mining technology in intrusion detection system, can draw the user's behavior characteristic, summarize the law of intrusion through analyzing historical data, Thus set up the more complete rules storehouse to intrusion detection. On the basis of analyzing the intrusion detection system existing, this paper has proposed an intrusion detection system based on classification rules. The intrusion detection system had higher measuring rate, better adaptive capacity and better expansibility.