Data received from billions of data points each day aggregating to Peta Bytes and from multiple network vendors – Nokia, Ericsson, Lucent. Needed real-time analytics to improve operational efficiency and reduce the Time to Repair.
Built a Big data analytics framework using the Apache stack.
Kafka for stream processing, Hadoop HDFS for storage, Map/Reduce jobs, Apache Spark, Python for ML and Tableau for visualization.
Used proprietary analytics framework built on Python for prediction using Machine learning.
Identified patterns from equipment data and characterized failure symptoms. Built a failure prevention model that predicts potential failures in advance with prescribed actions based on past experience. Average of 76% accuracy. This process improvement is expected to reduce operational costs by around 18% in the first year.