基于子类聚类和SAX表示的Shapelet快速发现算法

Shapelet发现的目标是寻找质量最佳的Shapelet，Shapelet的质量取决于子序列的可辨别性。针对精准发现有效Shapelet的问题，提出基于子类聚类和SAX表示的Shapelet快速发现算法，将子类聚类与经典的符号表示SAX法相结合进而快速准确的获取最优的Shapelet。该算法利用子类聚类将时间序列进行降维，得到多个子序列原型作为Shapelet候选集；再利用SAX表示将候选集符号化表示，直观的将候选集用字符串表示，便于找到最优Shapelet；最后选取候选集中信息增益最大的作为最优Shapelet进行时间序列分类。实验结果表明，该算法具有较好分类效果，同时提高了分类速度。

Fast Discovery Algorithm for Shapelet Based on Subclass Clustering and SAX Representation

The goal of Shapelet discovery is to find the Shapelet with the best quality, and the quality of Shapelet depends on the discriminability of the subsequence. Aiming at the problem of accurately discovering effective Shapelet, a fast discovery algorithm of Shapelet based on subclass clustering and SAX representation is proposed, which combines subclass clustering with classical symbolic representation SAX method to obtain the optimal Shapelet quickly and accurately. The algorithm uses subclass clustering to reduce the dimension of time series and obtains multiple subsequence prototypes as Shapelet candidate sets. SAX representation is used to symbolize the candidate set, and the candidate set is intuitively represented by string, which is convenient to find the optimal Shapelet.Finally, the Shapelet with the largest information gain in the candidate set was selected as the optimal Shapelet for time series classification. Experimental results show that this algorithm has good classification effect and improves the classification speed.