Classification and routing of IT incidents to the proper domain expert team is a key issue in the IT service industry. Existing IT service management systems involves manual classification and routing of incidents. Manual classification of incidents may involve misclassification and hence results in assigning of the incidents to a wrong resolver group. Combination of natural language processing and supervised machine learning techniques can be leveraged to develop an automated Incident categorization system using a labelled training data. This research work proposes a methodology to develop an automated incident classifier based on deep learning using LSTM- RNN models by mining the natural language incident description entered by the user. LSTM deep learning architectures are effective in memorizing important information and efficient in learning the long-term contextual dependencies exists in the incident description. Word embedding representations are used in this research work to numerically encode the incident descriptions. The efficacy of the proposed incident classifier model is empirically validated using a real IT infrastructure incident data and compared the results with Support Vector machines, Logistic Regression, Naive Bayes and K-Nearest Neighbour. The study showed that LSTM-RNN model with optimal hyperparameters reported the best performance results based on various classifier performance measures. Proper assignment of incidents to the respective domain team, speedy resolution, improved productivity, increased customer satisfaction and uninterrupted business are some of the advantages of the proposed automated incident classifier model.

IT incident management, Machine Learning, Natural Language Processing, Long Short-Term Memory, Recurrent Neural Networks, Word Embeddings.