Time Series AutoFeature Engineering: Fast, Scalable Method to Create Machine Learning-based Predictive Analytics from Time Series Data
Montag, 18. November 2019
Machine Learning applied on time series data arising from sensors have become a critical method to develop predictive analytics in industrial settings. A machine learning engineer spends most of his/her time in creating features from the raw time series data which is then fed to the machine learning algorithm. This features are either statistical, domain based or both. In this presentation, Tapan will present some of the novel methods GE Research has invented to automate the feature extraction for time series data. The key takeaway for the audience will be understanding subtleties of time series data, especially in a IOT scenario; abstraction of type of features for time series data; and two case studies of application of the method for building fast, scalable ML solutions are shown.