Abstract
This paper proposes an efficient method for clustering large amount of time sequences using dynamic time warping (DTW). This method introduces a "maxDist" constraint to limit the distance between time sequences. Based on this constraint, we developed a new hierarchical clustering algorithm which leverages multi-dimensional indexing to prune unnecessary DTW computations. In addition, in order to reduce the expensive computation cost of DTW, a new constraint-based DTW algorithm is proposed which can stop calculation for cells in the cost matrix once the partial distance is larger than the maxDist constraint. Experiments have been performed on a synthetic data set to evaluate its performance as well as on a real cyber security data set in order to identify groups of entities with common time sequence patterns.