Linear Fuzzy Information Granulation Based Classification Method for Unequal Length Time Series

For the classification of unequal length time series, the Dynamic Time Warping distance based k-Nearest Neighbors algorithm is one of the effective classification methods. However, this method has high time cost. Some existing methods decreased the time cost by transforming time series to granular time series, but the fuzzy information granules used there do not reflect the trend information of the given time series. The lack of trend information will lead to inaccurate classification results. In order to reflect the trend information, we adopt linear fuzzy information granules. Based on this, a new distance between granular time series, is defined for the corresponding time series, which can overcome the shortcomings of the Dynamic Time Warping algorithm. The proposed classification method built up on the newly defined distance exhibits better performance and higher efficiency in the experiments presented in this paper.