Financial time series indexing based on low resolution clustering

‡ Corresponding author Abstract One of the major tasks in time series database application is time series query. Time series data is always exist in large data size and high dimensionality. However, different from traditional data, it is impossible to index the time series in traditional database system. Moreover, time series with different lengths always coexists in the same database. Therefore, development of a time series indexing approach is of fundamental importance for maintaining an acceptable speed for time series query. By identifying the perceptually important points (PIPs) from the time domain, time series of different lengths can be compared and the dimensionality of the time series can be greatly reduced. In this paper, a time series indexing approach, based on clustering the time series data in low resolution, is proposed. This approach is customized for stock time series to cater for its unique behaviors. It follows the time domain approach to carry out the indexing process which is intuitive to ordinary data analysts. One may find it particularly attractive in applications like stock data analysis. The proposed approach is efficient and effective as well. As demonstrated by the experiments, the proposed approach speeds up the time series query process while it also guarantees no false dismissals. In addition, the proposed approach can handle the problem of updating new entries to the database without any difficulty.