AUTOMATION OF INCIDENT PRIORITIZATION IN IT SERVICE MANAGEMENT BASED ON ITIL

• Main Author: Angela, Tessa
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70707
• Institution: Institut Teknologi Bandung

Along with the rapid development of IT, the application of IT-based system has become more widespread, dramatically increasing the volume of IT incidents. Unfortunately, incident management practices, especially incident prioritization, is still done manually, making it prone against human errors and wasting a lot of time. Often time, determining the priority of an incident is done repeatedly due to the limited knowledge the initial level of support has, who generally is a Service Desk (SD). This results in inaccuracies, which the next level of support must correct. In addition, a formal feedback system relating to these errors is rarely found in companies, hampering the learning process of SD and thus, the exact problem keeps reoccurring. All these predicaments cause valuable resources to be redundantly wasted, thereby increasing the company’s expenditure unnecessarily. To overcome these problems, automation of IT incident prioritization is proposed. It is done by implementing the Information Technology Infrastructure Library (ITIL) as well as utilizing text mining, Natural Language Processing (NLP) and Machine Learning (ML). It is expected to increase the speed, consistency and learning curve in IT incident prioritization. The automation is carried out in several processes, namely attribute determination, text pre-processing, feature extraction and selection, imbalance data handling, prioritization based on urgency and impact levels, as well as user feedback involvement. First, attribute determination is done by utilizing textual description in the forms of summaries and descriptions listed on incident tickets. The next thing to do is to carry out the text pre-processing, consisting of case folding, tokenization, filtering by removing stop-words and stemming. Meanwhile, feature extraction and selection are performed using the Term Frequency, Inverse Document Frequency (TF-IDF) technique. Then, imbalance data handling is carried out using the oversampling method. Subsequently, the supervised learning method is done through Support Vector Machine (SVM), Decision Tree (DT), and Naïve Bayes (NB) to determine the level of urgency, impact and priority of the incident. Finally, the cross validation is performed using the k-Fold technique while the hyperparameter optimization is done by using the grid search technique. Should the results of determining the urgency, impact, and priority of the incident be considered inaccurate, users may revise them before the data is included into the training data set for the next learning session. iv In order to find out the performance improvement produced by the automation, a comparison is made between the performance of the automated system and of the manual system performed at a company being a case study in this research. Speed is measured from the moment an incident ticket is received by SD to read until the priority level of the incident is generated, while consistency is measured based on the prioritization results of the same number of incidents occurring at different times. Meanwhile, the learning rate is assessed based on the F1-score generated by the system being built and new SDs. Based on the implementation of all the stages mentioned above, it is evident that the automation of IT incident prioritization can be done and offers a better performance for incident management. The system development is carried out systematically by implementing ITIL, which provides an incident prioritization matrix based on impact and urgency levels. In addition, automation of the system is done by utilizing text mining, NLP, and ML. SVM turns out to give the highest F1-score among the three classifiers. Furthermore, the incident prioritization model is uploaded on a server and can be accessed as a web service for ticket management systems such as Jira SD. SD in-charge can also revise the level of impact, urgency, and priority of the incident produced by the automated system. The revised results are used as an input for further model trainings in order to improve the accuracy of the model in incident prioritization. Moreover, the automation increases the speed, consistency, and learning curve in IT incident prioritization. Based on the test results of the two IT staff and the automated system, the speed, consistency, and learning rate of the automated system is much better than the manual system being performed by the company currently.