Fault alarm data emanated from heterogeneous telecommunication network services and infrastructures are exploding with network expansions. Managing and tracking the alarms with trouble tickets using manual or expert rule-based methods have become challenging due to increase in the complexity of alarm management systems and demand for deployment of highly trained experts. As the size and complexity of networks hike immensely, identifying semantically identical alarms, generated from heterogeneous network elements from diverse vendors, with data-driven methodologies, has become imperative to enhance efficiency. In this article, data-driven trouble ticket prediction models are proposed to leverage alarm management systems. To improve performance, feature extraction, using a sliding time window and feature engineering, from related history alarm streams, is also introduced. The models were trained and validated with a data set provided by the largest telecommunication provider in Italy. The experimental results showed the promising efficacy of the proposed approach in suppressing false positive alarms with trouble ticket prediction.