Predictable packet filtering based on decision tree classifiers

Traditional technique for packet filtering is linear searching, in such way, filter efficiency is very low in the worst condition and rules in the rule-list are independent so that information among them can not be used effectively. In this paper, a new idea using a decision tree classifier is proposed. It first analyses the differences between packet filtering and general classification problems. Then the building and searching algorithms based on decision tree are explicated and the algorithm´s performance is analyzed. At last, the weight-threshold method is proposed to make the classifier predictable and reduce the matching workload. It can be illustrated that packet filtering based on decision tree classifiers is more efficient and can study inductively from the rules.