ICTACT Journal on Soft Computing

S. P. Paramesh ¹, K. S. Shreedhara ¹

Affiliations
1 Department of Computer Science and Engineering, University B.D.T College of Engineering, IN

Abstract

Proper assignment of IT incident tickets raised by the end users is a very crucial step in an IT Service management system. Incorrect manual selection of incident category while raising the ticket causes assignment of incident to a wrong domain expert team which in turn results in unnecessary resolution delay and resource utilization. In this work, we proposed machine learning based model for auto categorization of incident category by mining the user’s natural language description of the incident. Classification techniques such as Naive Bayes and Support Vector Machines are used as base classifiers to model the incident classifier system. To further analyse the classifier performance we used the ensemble classifier techniques such as Bagging and Boosting to build the incident classifier model. The performance of base classifiers and ensemble of classifiers are analysed using various performance metrics. Ensemble of classifiers outperformed well in comparison with the corresponding base classifiers. Pre-processing of the IT incidents description data is one of the key challenges in this research work due to its unstructured nature. The proposed automated incident classification model results in simplified user interface, faster resolution time, improved productivity and user satisfaction and uninterrupted flow in business operations. The real world IT infrastructure incidents data from a reputed enterprise is used for our research purpose.

Keywords

Machine Learning, Incident Classification, Ensemble Classifiers, Naive Bayes, Support Vector Machine.

Full Text

     

S. P. Paramesh ¹, K. S. Shreedhara ²

Affiliations
1 Department of Computer Science and Engineering, School of Engineering, Central University of Karnataka, IN
2 Department of Studies in Computer Science and Engineering, University B.D.T College of Engineering, IN

Abstract

Assignment of problem tickets to a proper resolver group is an important aspect and crucial step in any IT Service management tools like IT Service desk systems. Manual categorization of tickets may lead to dispatching of problem tickets to an inappropriate expert group, reassignment of tickets, delays the response time and interrupts the normal functioning of the business. Traditional supervised machine learning approaches can be leveraged to train an automated service desk ticket classifier by using the historical ticket data. Sparsity, non-linearity, overfitting and handcrafting of features are some of the issues concerning the traditional ticket classifiers. In this research work, a deep neural network based on Convolution Neural Network (CNN) is proposed for the automated classification of service desk tickets. CNN automatically extracts the most salient features of the ticket descriptions represented using word embeddings. The extracted features are further used by the output classification layer for efficient ticket category prediction. To corroborate the efficacy of the proposed ticket classifier model, we empirically validated it using a real IT infrastructure service desk data and compared the results with the traditional classifier models like Support Vector machines, Naive Bayes, Logistic Regression and K-nearest neighbour. The proposed CNN model with proper hyperparameters tuning outperforms the traditional classifiers in terms of overall model performance. Assignment of tickets to the correct domain groups, speedy resolution, improved productivity, increased customer satisfaction and uninterrupted business are some of the benefits of the proposed automated ticket classifier model.

Keywords

Service desk, Machine learning, Deep neural networks, Convolution Neural Network, Word Embeddings

Full Text