Threshold Similarity Queries in Large Time Series Databases

Abstract

Similarity search in time series data is an active area of research. In this paper, we introduce the novel concept of threshold-similarity queries in time series databases which report those time series exceeding a user-defined query threshold at similar time frames compared to the query time series. In addition, we present a new data structure to support threshold similarity queries efficiently. The performance of our solution is demonstrated by an extensive experimental evaluation.

Periodic Pattern Analysis in Time Series Databases

“…As proposed in [3,4], time series considering a single time domain can be represented as a sequence of intervals according to a certain threshold value τ . For the recognition of relevant periodic patterns that are hidden in the matrix representation of dual-domain time series, we extend this approach to a novel abstract meaning.…”

“…Considering the single-domain representation of time series, we performed similarity queries on the given datasets utilizing the techniques that are applicable for computing similarity on single-domain time series, such as the Euclidean distance (in the following denoted as EUCL), the DTW [7] and the threshold-based approach [3,4], in the following referred to as THR. Later in this section, we outline the obtained results of our newly introduced approach of measuring the distances for comparison.…”

“…Here, we take special emphasis on events that refer to an exceeding of a given activity threshold. Similarity search methods based on events that refer to exceeding of a given activity threshold have been introduced in [3,4]. Given a certain threshold value τ , this approach reduces singledomain time series to a sequence of intervals corresponding to time periods where the amplitude value of a time series exceeds τ .…”

TiP: Analyzing Periodic Time Series Patterns

“…An intersection set P τ (X) of a time series X is created by a set of polygons that are derived according to [2], where time series with one time domain are represented as a sequence of intervals w.r.t. to a threshold value τ .…”

“…Beyond the interest of [2], the temporal location and the evolution of consecutive patterns are focused. For efficient similarity computation, relevant feature information is extracted so that data mining techniques can apply.…”