Predictive Analytics of Sensor Data Using Distributed Machine Learning Techniques

Abstract

This work is based on a real-life data-set collected from sensors that monitor drilling processes and equipment in an oil and gas company. The sensor data stream-in at an interval of one second, which is equivalent to 86400 rows of data per day. After studying state-of-the-art Big Data analytics tools including Mahout, RHadoop and Spark, we chose Ox data's H2O for this particular problem because of its fast in-memory processing, strong machine learning engine, and ease of use. Accurate predictive analytics of big sensor data can be used to estimate missed values, or to replace incorrect readings due malfunctioning sensors or broken communication channel. It can also be used to anticipate situations that help in various decision makings, including maintenance planning and operation.