BOOT-TS: A Scalable Bootstrap for Massive Time-Series Data

We propose a scalable method of assessing the quality of machine learning algorithms over sampled time-series data. While bootstrap provides a simple and powerful means of estimating accuracy, its application to large time-series data still suffers from scalability issues. As an alternative we introduce BOOT-TS, a scalable extension of bootstrap for time-series which utilizes the recent advances in bootstrap and time-series theory to provide a practical implementation for assessing a time-series sample quality using Hadoop. For instance, our new procedure yields a robust and computationally efficient means of assessing the quality of our Twitter analytics workflow over large, real-world, time-series data.