PLA data reduction for speeding up time series comparison

Abstract EN

We consider comparison of two Piecewise Linear Approximation (PLA) data reduction methods, a recursive PLA-segmentation technique (Douglas-Pucker Algorithm) and a sequential PLA-segmentation technique (FAN) when applied in prior of our previously developed time series alignment technique SEA, which was established as a very effective method.

The outcome of these two combination are two new time series alignment methods : Rec SEA and Seq SEA.

The study shows that both Rec SEA and Seq SEA perform alignments as good as those of SEA with important reductions in data (Rec SEA: up to 60 %, Seq SEA up to 80 % samples reduction) and processing time(Rec SEA: up to 85 %, Seq SEA up to 95% time reduction) with respect to the SEA method.

This makes both the two new methods more suitable for time series databases querying, searching and retrieval.

Particularly, Seq SEA is significantly much faster than Rec SEA for long time series.