Exploiting partial decision trees for feature subset selection in e-mail categorization

In this paper we propose PARTfs which adopts a supervised machine learning algorithm, namely partial decision trees, as a method for feature subset selection. In particular, it is shown that an aggressive reduction of the feature space can be achieved with PARTfs while still allowing for comparable classification results with conventional feature selection metrics. The approach is empirically verified by employing two different document representations and four different text classification algorithms that are applied to a document collection consisting of personal e-mail messages. The results show that a reduction of the feature space in the magnitude of ten is achievable without loss of classification accuracy. Categories and Subject Descriptors H.3.3 [Information Systems]: Information Search and Retrieval; H.3.1 [Information Systems]: Content Analysis and Indexing General Terms Indexing methods, information filtering, feature selection Keywords Text categorization, machine l...

• Type: Conference
• Year: 2006
• Where: SAC
• Authors: Helmut Berger, Dieter Merkl, Michael Dittenbach
• Keywords: Applied Computing, Feature Selection, Feature Space, Feature Subset Selection, SAC 2006