Fast algorithm for mining similar sub-trend sequences in time-series databases
Si Guo · Journal of Zhejiang University(Engineering Science) · 2002
Mining similar trend sequences is a novel problem in the data mining field. Definitions of trend similarity and trend indicator distribution were given. Properties of trend indicator distribution and its relationship with trend similarity were studied. Results showed that trend indicator distribution could be applied to discard most unsimilar candidate sub trends before the trend matching. Besides, the search could be carried out in a jumping manner. Implementing these two ideas, a fast algorithm, VISL(Variable Incremental Step Length), was proposed for mining sub trend sequences similar to the given query trend in a long related. Performance in running time of VISL was tested against two other algorithms, one of a running time independent of the degree of similarity, and the other being relatively fast among algorithms available in the literature. A better running time performance was achieved by VISL in a relatively high user defined degree of similarity environment. Space cost of the algorithm was also discussed.