A multilevel adaptive reduction technique for time series
Hamdi Yahyaoui, Hosam M. F. AboElFotoh, Yanjun Shu · Neurocomputing · 2023
We devise in this paper a Multilevel Adaptive Reduction Technique (MART) for time series data. MART extracts the main features of a time series and encodes them in a reduced data structure called reduct. The extracted features are leveraged to establish a distance between reducts that satisfies the lower bounding constraint. Furthermore, we show how MART can be further applied on the reduced data and operate at different levels. We conduct experiments on time series datasets that show MART based classification accuracy and multilevel reduction power. We present also a comparative study with well-known SAX based time series reduction techniques and deep learning time series classification techniques.