Adaptive Multiresolution and Dedicated Elastic Matching in Linear Time Complexity for Time Series Data Mining

We develop an adaptive multiresolution approach to the problem of multidimensional time series characterization. Furthermore we provide a dedicated elastic pseudo distance to support similarity search mechanisms for such characterization. We show theoretically and experimentally that our multiresolution decomposition of times series has a linear complexity in time and space. The pseudo elastic distance AMR-DTW that we develop to match multiresolution representations of time series is also based on iterative algorithms that show linear time and space complexity for some tuned parameters. We evaluate the proposed adaptive multiresolution algorithm and associated pseudo elastic distance in a classification experiments to demonstrate the efficiency and accuracy of the proposed representation and matching scheme for time series data mining