| Summary: | 碩士 === 國立中山大學 === 資訊工程學系研究所 === 103 === In recent years, time-series data mining has been considered as an important topic attracting many researchers. One of the most important topics in the time-series databases is to find the periodicity of the patterns. Periodic pattern mining is useful in predicting the stock price movement, computer network fault analysis and detection of security breach, earth-quake prediction, and gene expression analysis. It is difficult because it not only needs to use the information in the time-series database to find out the frequent patterns, but also needs to make sure the patterns which are frequent patterns occurring in the similar period length. Therefore, a new concept of finding time-series periodic patterns is proposed by Nishi et al., which cares about the patterns that are frequent for a flexible period of time from the time-series database. Nishi et al. also states the concept to define the flexible period patterns. However, the algorithm proposed by Nishi et al. has some problems for finding the user interesting patterns. When they derive frequent periodic 1-patterns, they need many times to store all the patterns which is frequent in the array. Moreover, when generating candidate periodic k-pattern ( k≧2 ), Nishi et al.''s algorithm may check all candidate periodic k-patterns instead of focus on generating the user interesting patterns. It also wastes execution time. Therefore, to avoid these problems and improve the performance, we propose a Restriction-Based algorithm to efficiently find out the user interesting patterns. We present the pruning strategies during deriving frequent periodic 1-patterns. These pruning strategies not only can be applied to check whether the items are frequent periodic 1-patterns or not but also satisfy the restriction for the generating the candidate patterns. The strategy could reduce the execution time.
Furthermore, we also propose a join policy to focus on generating the user interesting patterns. Therefore, our algorithm can avoid getting unwanted results. From our simulation results, we show that our Restriction-Based algorithm is more efficient than Nishi et al.''s algorithm.
|
|---|
