Scalable Time Series Classification in streaming and batch environments on Apache Spark

Time series classification is an important problem since data from sensors become more prevalent over time. In addition most of the data arrive in the form of a stream and thus have to be handled with the limitation that apply to streaming environments (low latency,low memory footprint). In this paper we address the problem of scalable time series classification on both Batch and Streaming environments. More specifically we implemented two state-of-the-art time series classification on top of Apache Spark and we adapted one of them for streaming applications. We evaluated our algorithms against two open datasets on a 10-node cluster. The algorithms we implemented scaled gracefully both in the batch and streaming environment.