Boundary-Based Time Series Sorting

In many applications, it is desirable to sort the data. Most of previous work on sorting are key based, however, there are no apparent keys for the time-series data and therefore the classic sorting algorithms may fail in sorting time-series data. We propose a novel technique, called TS-Sort, to sort time-series sequences in the massive set. The proposed method first extracts the maximum and minimum boundaries of the set, then calculates the distance values between the sequences to the boundaries, and finally sorts the values to determine the relative orders of sequences in the set. For improvement, we propose a partition based version of the algorithm, which puts the sequences into small groups, and sorts the groups to get the final sorted set. Extensive experiments, both on synthetic and real datasets, show that our approach can be used to make the time series set in order, and there is a factor of up to 26.3% accelerating for the improved version of the method.