Business process forecasting in telecom industry

Telecommunication companies deal with thousands of business processes daily. As different tasks progress along the process sequence the process owners are interested in three key questions: what other events might still occur, when and how will the process end? All these questions are addressed here in an attempt to create a generic framework for predicting various properties of the remaining process flow based on the data extracted from a combination of live and historical processes. The proposed methodology exploits both categorical and numerical features extracted at the process and individual tasks levels to deliver some predictive characteristics of the remaining process path. In the example studied in this paper the focus is on predicting the remaining fault repair process carried out on telecommunication lines. A set of methods ranging from Markov chains up to decision trees are applied and tested, showing some interesting results and hinting for further research.