A Comparison of Text Classifiers on IT Incidents Using WEKA

Abstract

IT service management and incident management is a hot topic in every company which serves IT services and they require human effort to manage. In ITIL framework for IT service management, it's always useful to link the incidents with configuration items, in other words the assets or components necessary to deliver IT services and this task is managed manually by IT support technicians in many IT service management tools. The aim of the study is to remove this manual linking step by applying text classification methods and instead to provide an automatic assignment of CI's to incidents. Four of the text classification methods, Naive Bayes Multinomial, k-Nearest Neighbor, Support Vector Machine and J48 decision tree classifiers, are used on three different sets of incidents extracted from the same database by using different filters. The impact of some pre-processing steps is compared on different sizes of datasets.