Time Series Classification Using Non-Parametric Statistics

We present a new class-based prediction algorithm for time series. Given time series produced by different underlying generating processes, the algorithm predicts future time series values based on past time series values for each generator. Unlike many algorithms, this algorithm predicts a distribution over future values. This prediction forms the basis for labelling part of a time series with the underlying generator that created it given some labelled examples. The algorithm is robust to a wide variety of possible types of changes in signals including mean shifts, amplitude changes, noise changes, period changes, and changes in signal shape. We show results demonstrating that the algorithm successfully segments signals for a wide variety of example possible signal changes.