A global telecom hardware vendor was looking to add cutting-edge anomaly detection in their products for mobile service providers. The objective was to enable the providers to detect unusual or abnormal occurrences or bursts in their networks. The network data that we had to work with had its challenges: there were terabytes of unstructured, real-time, non-stationary time-series data coming from various sources every day, as well as an absence of previously labeled data, plus thousands of parameters requiring analysis.
Given the nature and volume of the data we focused on unsupervised machine-learning techniques as the means to enable the system to “learn” from that data without human input or instruction. What was it that it learned? It learned to instantly identify and flag abnormal events in mobile traffic. In this effort, we tested multiple algorithms before finalizing our approach utilizing Bayesian Networks and Hidden Markov Model in formulating our solution. The end result was what the client expressly wanted: a feature in their products that could detect anomalies in real-time and in real-world network traffic.
This project was a particularly demanding challenge. We worked closely with the client’s engineers to understand the intricacies of mobile networks before embarking on our solution. In addition to developing the detection capability, we also set up a Big Data infrastructure capable of handling extremely large amounts of streaming real-time data (tens of megabytes per second). We also developed a solution for traffic-load prediction and resource planning using regression models.
The solution we devised for our client’s products gave them a world-class anomaly detection capability and an improved strategic position in the marketplace. This led to increased revenue, customer satisfaction, new customers, and fulfillment of their business goals.