TMCnet News
Machine Learning does the (previously) impossible: stopping fraud in its tracks in wholesale telco carrier networksSAN MATEO, Calif., Oct. 24, 2017 (GLOBE NEWSWIRE) -- Argyle Data has demonstrated the ground-breaking capabilities of its machine learning-based predictive analytics for two European wholesale carriers in U.S. and European lab tests. According to recent reports by industry analysis firm ABI Research, machine learning-based predictive analytics are applicable to all aspects of the telecom business, and these technologies will lead operators to profoundly change how they manage the telecom business. ABI estimates that mobile operators will devote more than U.S.$50 billion to big data and machine learning through 2021. Two-thirds of all carrier fraud losses are related to international traffic, and much of this traffic passes through wholesale networks. Mobile broadband operators worldwide have made major investments to develop internal fraud detection and prevention systems. However, the wide variations in technology, internal processes and national regulations make it difficult for operators to orchestrate fraud defenses beyond their own systems. For wholesale carriers, the task is exponentially tougher. Wholesalers typically receive only minimal data about their retail telecoms customers’ traffic – too little for rules and thresholds-based detection systems and too late for any action other than historical analysis. In an increasingly competitive and tight-margin industry, wholesale carriers are seeking new ways to provide added value to their retail customers and affiliates. With international scams the highest contributors to mobile fraud losses, wholesalers are increasingly looking to predictive analytics to stop fraud as it occurs, regardless of the origin or destination of the suspect traffic. 1. Top tier French operator The operator provided two weeks of historical data for their proof of concept trial in Argyle Data’s U.S. laboratories. A Supervised Machine Learning algorithm approach was used, leveraging a Hadoop architecture to create a data lake that eliminated the isolation of data into silos. Even with this limited data, Argyle Data:
2. European Communications Hub Provider One month of very limited historical data was provided. This was analyzed in the local country laboratories for reasons of data security. Again, a Supervised Machine Learning algorithm approach was used, based on a Hadoop data lake architecture. Even with this severely restricted data, results included:
Argyle Data VP of Engineering Padraig Stapleton said, “Initially, each of the wholesalers doubted whether a machine learning approach could be applied to wholesale traffic routing, because they were all too aware of the hugely varied amount of traffic crossing their networks and large gaps in traffic flow from different customers at different times. The fear was that the machine learning system could not learn what was normal in order to create a baseline for detecting abnormalities. Both operators were totally convinced after seeing our results, and our work with them is ongoing. It’s hard to argue with an unprecedented achievement in fraud detection at the wholesale level.” More information on these case studies is available at www.argyledata.com/resources. About Argyle Data™ Contact: Mary McEvoy Carroll Argyle Data [email protected] + 1-408-691-4283 |