Research on time series data mining based on linguistic concept tree technique

Mining qualitative predictive knowledge for time-series has been listed as one of the challenge for Time Series Data Mining. Euclidean distance was used extensively. However, it was a brittle distance measure because of less robustness. A modification algorithm used linguistic variable concept tree to describe the slope feather of time series. For reducing the computational time and local shape overwhelming, the piecewise linear representation was used to preprocess original series. In addition, the linguistic variable replaced the slope of piecewise series in time warping, which can compensate dismissing of important feature as linear reduction in preprocessing, moreover different time granularity analysis can also be executed being the linguistic variable tree constructed easily. All the method was based on cloud models theory which integrities randomness and probability of uncertainty. Experiment results show this method has strong robustness and more accurate for time series analysis.