Piecewise cloud approximation for time series mining

Many researchers focus on dimensionality reduction techniques for the efficient data mining in large time series database. Meanwhile, corresponding distance measures are provided for describing the relationships between two different time series in reduced space. In this paper, we propose a novel approach which we call piecewise cloud approximation (PWCA) to reduce the dimensionality of time series. This representation not only allows dimensionality reduction but also gives a new way to measure the similarity between time series well. Cloud, a qualitative and quantitative transformation model, is used to describe the features of subsequences of time series. Furthermore, a new way to measure the similarity between two cloud models is defined by an overlapping area of their own expectation curves. We demonstrate the performance of the proposed representation and similarity measure used in time series mining tasks, including clustering, classification and similarity search. The results of experiments indicate that PWCA is an effective representation for time series mining.