How to Quantify the Impact of Lossy Transformations on Event Detection

  • Author:

    Pavel Efros, Erik Buchmann, Adrian Englhardt, Klemens Böhm

  • Source:

    Big Data Research, special issue on Online Forecasting and Proactive Analytics in the Big Data Era, Elsevier Science Publisher, 2017

  • Abstract

    To ease the proliferation of big data, it frequently is transformed, be it by compression, be it by anonymization. Such transformations however modify characteristics of the data. In the case of time series, important characteristics are the occurrence of certain changes or patterns in the data, also referred to as events. Clearly, the less transformations modify events, the better for subsequent analyses. More specifically, the severity of those modifications depends on the application scenario, and quantifying it is far from trivial. In this paper, we propose MILTON, a flexible and robust Measure for quantifying the Impact of Lossy Transformations on subsequent event detectiON. MILTON is applicable to any lossy transformation technique on time-series data and to any general-purpose event- detection approach. We have evaluated it with several real-world use cases. Our evaluation shows that MILTON allows to quantify the impact of lossy transformations and to choose the best one from a class of transformation techniques for a given application scenario.


    We provide the code for MILTON: Download archive.
    The code contains an example of transformed and original time series.


    download pdf