Feature-Based Time Series Classification for Service Request Opening Prediction in the Telecom Industry

Telecommunication companies face the challenge to reduce the number of service request openings (SROs). A predictive behavior able to reduce this number can improve customers experience and decrease operational costs. This paper proposes a machine learning (ML) based approach to reduce the number of SROs. For such, it uses real data from a Brazilian telecom operator. The proposed approach uses feature-based time series extracted from network equipment’s signals, modeling the problem as a binary classification task. We carry out experiments to investigate the impact of long-term and short-term windows in the predictive performance. After pre-processing the data, we apply different classifiers algorithms. According to experimental results, a high predictive performance was obtained, mainly when long-term network behavior data was used. These results have a positive impact in the company costs.