A fast algorithm for mining similar trends in trend-sequence databases

Mining time series data is a novel and important issue in the field of data mining. An approach of retrieving interesting information in the time series is proposed. This approach converts time series into symbolized, "trend sequences", and then considers the task of finding the trends similar to given query trend in the trend databases. To solve the problem, definitions of trend are given, consequently followed by definitions and properties of trend distribution. An index based on trend distribution is constructed and applied for filtering out dissimilar candidate trends to accelerate searching. Experiments show that this index is effective to accelerate the searching of similar trends, under the condition of small trend indicator set and trends being within "low frequency" band.