Comparison of rule based classification techniques for the Arabic textual data

Abstract

Text categorisation discipline has recently attracted many scholars because of the large number of documents on the World Wide Web (WWW) that contain hidden useful information which can be utilised by organisational's managers for decision making. However, the majority of research conducted in text categorisation is related to English data collections while there is limited research attempts conducted on mining corpuses in Arabic. This paper investigates the problem of Arabic text categorisation in order to measure the performance of different rule based classification data mining techniques. Precisely, four different rule based classification approaches: C4.5, RIPPER, PART, and OneRule are compared against the known CCA Arabic text data set. Experiments are carried out using a modified version of WEKA business intelligence tool, and the results determine that the least suitable classification algorithms for classifying Arabic texts is OneRule whereas RIPPER, C4.5 and PART have similar performance with respect to error rate.