Hybrid Forecasting of Service Request Tickets with Human-in-the-loop (HitL)
Time-series forecasting is a well explored topic in machine learning. However, in some forecasting applications, time-series models like LSTM, ARIMA, VARMAX, etc. are not able to capture all the relevant features and external factors that may influence its prediction. Additionally, when dealing with high dimensionality data, the task of extracting important and discriminative features is non-trivial. We present a novel application in the services industry where we developed a time-series forecasting model with prophet  and augmented HitL to improve its prediction. Previous methods [3,4] suffer from not having factors that account human feedback, diversity in scenarios in the same dataset, data features, etc. Our method leverages the HitL to augment this time-series model and navigate with better features to get the best-case scenario with optimal performance. Our novel method is applied to an application of a global IT service provider, which has thousands of product portfolios that sells to enterprise clients.