Novel fuzzy classification approaches based on optimisation of association rules

The present paper proposes an approach for classification based on fuzzy rules. The paper mainly concentrates to optimize the association rules for classification. The present study proposes a method called Integrated Rule Classifier (IRC). To develop a Fuzzy Association Rule (FAR) algorithm to produce rules which are suitable for signature-based and anomaly-based detection for mining purposes with attacks. The proposed IRC derives cluster representatives. In the second step FAR's are formed. A lot of flexibility is achieved by the proposed IRC scheme which is not possible in the existing algorithms. One can use any clustering algorithm in step-1 depending on the data-set and other constraints. The other flexibility is that FAR's can be formed based on all cluster representatives or randomly chosen representatives. The proposed IRC methodology is experimented on network audit data collected from KDDCUP99 data-set with class-labels of various network attacks.