Building a classification engine for ticket routing in IT support systems

In any IT support environment, it is important to quickly route support tickets to correct teams. Often, it takes few days to manually classify several hundreds of tickets. This thesis presents a classification engine that provides routing recommendation to specialists for incoming tickets. The classification engine is built using machine learning and software robotics to decrease the amount of human time spent on support ticket classification. Experiments are carried with logistic regression, random forests and extremely randomized trees using historical data. During off-line cross-validation, random forest model performs well with 90% of f1-score and is deployed in production using AWS. The performance of the classification engine is tested in production for two weeks. The deployed model has f1-score of 86%. The f1-scores for the individual groups like level 1, level 23, level 24 are 89%, 88% and 93% respectively. These three groups contribute to almost 90% of total tickets. This thesis presents an approach of how a machine learning model is employed to reduce human time