关键词: 时间序列分类, shapelets转换, 局部敏感哈希

Abstract:

Aiming at the problem that the time series classification algorithm based on shapelets conversion is too time-consuming due to the existence of a large number of similar sequences in the shapelets candidate set, a LSH-based Shapelets Transformation method（Locality Sensitive Hashing Shapelets Transform, LSHST） is proposed. An improved algorithm of Local Sensitive Hash（LSH） function is proposed, and the original subsequence candidate set is filtered step by step. The quality of shapelets in the set is calculated, and the amount of shapelets is reduced by a method of coverage to determine the shapelets to be transformed, and finally shapelets is transformed. The experiments show that, compared with algorithms such as Shapelet Transform（ST）, ClusterShapelet（CST）, and Fast Shapelet Selection（FSS）, LSHST improves the classification accuracy by 20.05, 19.95, and 16.52 percentage points, and the highest time savings 8,000 times, 16,000 times and 8.5 times.

Key words: time series classification, shapelets transform, Locality Sensitive Hashing（LSH）