MICRO-DISTANCE DISCOVERY IN TIME SERIES DATABASE

Time series constitute a large part of data stored in information system. Computing the distance of two time series is a crucial problem in many data mining application. The current time series association mining research is based on the overall series shape. But, in some areas, the local or intuitive series shape is not what can be ignored but what is of interest. For this reason, a new distance computing method for time series is proposed, which is named micro distance （ MD （X, Y） ）. And a new algorithm, called the FDD algorithm, is proposed, which maps time series to the frequency domain, disjoining different subseries pattern to determine the distance of two time series. The FDD algorithm is irrelative to base line and scale of series. And for the huge series, it is a fast algorithm.