Shape clustering on time series data

The trend is extracted from raw time series data by wavelet. The maximum points of the trend are found and considered as critical points of this time series. These critical points describe the shape of time series. Shape clustering is performed on all time series. This method is capable of clustering sequence and subsequence. Some specific shapes are extracted from time series. In stock technical analysis, some specific shapes (patterns) hidden in real time series, for example double bottom, head and shoulder top, are analyzed. The experiments were performed on stock time series data. The experimental results show that the shape clustering based on time series is effective.