THE AUTOMATED MACHINE LEARNING CLASSIFICATION APPROACH ON TELCO TROUBLE TICKET DATASET

This paper presents automated machine learning for solving a practical problem of a telco trouble ticket system. In particular, the paper's focus is on the classification of early resolution code from the trouble ticket dataset. The number of trouble tickets is rising every year due to the new challenges from the digital world. It is a challenging job to evaluate the vast content of the trouble tickets manually. Past trouble ticket contains essential information about the root cause and the resolution to each problem. The main contribution of providing the early resolution code for each new trouble ticket can significantly reduce Mean Time to Restore (MTTR) for the telco operation, thus improves customer satisfaction and minimize telco business and operation costs. The research methods include the existing traditional model and its modification towards the best accuracy. Automated Predictive Engine (AutoPE) improves the current traditional engineered model's classification accuracy from 5% to 38% when using the optimal performing solution by implementing AutoML and Grid Search. This solution uses multiple classifiers such as Random Forest, Deep Learning, Gradient Boosting, XGBoost, and Extremely Randomized Trees classifiers on a set of features based on various telco serviceable broadband zones and sampling size. Finally, compared to the baseline existing traditional engineered model, the best performing solution also improves the quality of classification for the early resolution code for the telco trouble tickets dataset.