Multi-dimensional Time Series Approximation Using Local Features at Thinned-out Keypoints

A multi-dimensional time-series is a sequence of vectors measured by many devices at points in time. Although many methods have been proposed to model and classify the data, these methods lead to a problematic relationship between cost and accuracy. In this paper, we propose a novel method for approximating multi-dimensional time-series, named multi-dimensional time-series Approximation with use of Local features at Thinned-out Keypoints (A-LTK), which enables an adequate accuracy value to be obtained even when reduced storage cost is a requirement. The main concepts of A-LTK are 1) reduction of time points and 2) construction of local features at the thinned-out keypoints. A preliminary evaluation indicated that with these points our proposed method was capable of achieving almost the same accuracy with less storage cost, compared to existing methods.