Multi-label rules for phishing classification

Abstract

Generating multi-label rules in associative classification (AC) from single label data sets is considered a challenging task making the number of existing algorithms for this task rare. Current AC algorithms produce only the largest frequency class connected with a rule in the training data set and discard all other classes even though these classes have data representation with the rule’s body. In this paper, we deal with the above problem by proposing an AC algorithm called Enhanced Multi-label Classifiers based Associative Classification (eMCAC). This algorithm discovers rules associated with a set of classes from single label data that other current AC algorithms are unable to induce. Furthermore, eMCAC minimises the number of extracted rules using a classifier building method. The proposed algorithm has been tested on a real world application data set related to website phishing and the results reveal that eMCAC’s accuracy is highly competitive if contrasted with other known AC and classic classification algorithms in data mining. Lastly, the experimental results show that our algorithm is able to derive new rules from the phishing data sets that end-users can exploit in decision making.