Alternative quality measures for time series shapelets

Classification is a very broad and prevalent topic of research within data mining. Whilst heavily related, time series classification (TSC) offers a more specific challenge. One of the most promising approaches proposed for TSC is time series shapelets. In this paper, we assess the current quality measure for shapelet extraction and introduce two statistical tests for shapelet finding. We show that when compared to information gain, these two quality measures can speed up shapelet extraction whilst still producing classifiers that are as accurate as the original.