Efficient Temporal Kernels Between Feature Sets for Time Series Classification

Abstract

In the time-series classification context, the majority of themost accurate core methods are based on the Bag-of-Words framework,in which sets of local features are first extracted from time series. Adictionary of words is then learned and each time series is finally repre-sented by a histogram of word occurrences. This representation inducesa loss of information due to the quantization of features into words asall the time series are represented using the same fixed dictionary. Inorder to overcome this issue, we introduce in this paper a kernel operat-ing directly on sets of features. Then, we extend it to a time-compliantkernel that allows one to take into account the temporal information. Weapply this kernel in the time series classification context. Proposed kernelhas a quadratic complexity with the size of input feature sets, which isproblematic when dealing with long time series. However, we show thatkernel approximation techniques can be used to define a good trade-offbetween accuracy and complexity. We experimentally demonstrate thatthe proposed kernel can significantly improve the performance of timeseries classification algorithms based on Bag-of-Words.