This work motivates the need for more flexible structural similarity measures between time-series sequences, which are based on the extraction of important periodic features. Specifically, we present non-parametric methods for accurate periodicity detection and we introduce new periodic distance measures for time-series sequences. We combine these new measures with an effective metric tree index structure for efficiently answering k-Nearest-Neighbor queries. The goal of these tools and techniques are to assist in detecting, monitoring and visualizing structural periodic changes. It is our belief that these methods can be directly applicable in the manufacturing industry for preventive maintenance and in the medical sciences for accurate classification and anomaly detection.

In: Data Mining Knowledge Discovery

Details