A Hybrid Machine Learning Framework of Gradient Boosting Decision Tree and Sequence Model for Predicting Escalation in Customer Support
Shubham Babu Gupta · 2020
Large software organizations spend millions of dollars on their technical support system to resolve their users' concerns, provide excellent quality of support, and reduce friction from their systems. Sometimes situations arise where the expectation of customer conflict with the support team process, which may lead to the escalation of support cases. An escalation causes a lot of problems for the company and they try every possible solution to avoid escalation. Machine learning solutions help organizations to provide a seamless experience to their customers and help them to remain competitive in the market. This paper aims to predict whether a particular support case will be escalated by a customer in the nearest future based on past escalation cases, metadata recorded when the case opened, and information generated through the conversation customers have on their support ticket with the company representatives. We have developed a hybrid modelling framework of XGBoost (gradient boosting) and LSMT (sequence) model to effectively use sequential textual information, time-series data, and metadata for modelling. The modelling framework is generic and can be extended to any problem where both metadata (non time-series) and time-series available. For model interpretation and understanding of the main drivers behind escalation, we used global and local feature interpretability of the SHapley Additive exPlanations (SHAP) approach. This paper is a response to the IEEE BigData 2020 Cup challenge.