ITSM automation - Using machine learning to predict incident resolution category

Abstract

Problem resolution is a key issue in the IT service industry, and it is still difficult for large enterprises to guarantee the service quality of the Incident Management (IM) process because of the difficulty in handling frequent incidents timely, even though IT Service Management (ITSM) standard process have already been established (Zhao & Yang, 2013). In this work, we propose an approach to predict the incident solution category, by exploring and combining the application of natural language processing techniques and machine learning algorithms on a real dataset from a large organization. The tickets contain information across a vast range of subjects from inside the organization with a vocabulary specific to these subjects. By exploring the text-based attributes, our findings show that the full description of an incident is better than the short description and after stop words removal, the use of additional preprocessing techniques and the addition of tickets nominal attributes such as have no impact to the classification performance.