SwiftRule: Mining Comprehensible Classification Rules for Time Series Analysis

Abstract

In this article, we provide a new technique for temporal data mining which is based on classification rules that can easily be understood by human domain experts. Basically, time series are decomposed into short segments, and short-term trends of the time series within the segments (e.g., average, slope, and curvature) are described by means of polynomial models. Then, the classifiers assess short sequences of trends in subsequent segments with their rule premises. The conclusions gradually assign an input to a class. As the classifier is a generative model of the processes from which the time series are assumed to originate, anomalies can be detected, too. Segmentation and piecewise polynomial modeling are done extremely fast in only one pass over the time series. Thus, the approach is applicable to problems with harsh timing constraints. We lay the theoretical foundations for this classifier, including a new distance measure for time series and a new technique to construct a dynamic classifier from a static one, and demonstrate its properties by means of various benchmark time series, for example, Lorenz attractor time series, energy consumption in a building, or ECG data.

All for one or one for all? Combining heterogeneous features for activity spotting

“…In [14], [15] we have shown that a 0 , a 1 , a 2 , a 3 , etc. can be interpreted as the optimal estimators of average, slope, curve, change of curve, etc.…”

“…With a given representation of a time series by orthogonal expansion coefficients, two time series can be compared simply by taking the Euclidian distance (or the scaled Euclidean distance) of two orthogonal expansion coefficient vectors [15]. Doing so, the temporal effort to compare precomputed coefficient vectors is marginal.…”

“…However, depending on the application we may also wish to compare two time series neglecting constant offsets in the target domain or different lengths. This can be done easily with some simple transformations as shown in [15].…”

“…In addition to the coefficients, the lengths of the time series themselves and the approximation errors turned out to be useful features in various applications [15]. In this paper, polynomial features are used for activity spotting for the first time.…”

Improved Gath–Geva clustering for fuzzy segmentation of hydrometeorological time series

“…This problem has been researched extensively and has not yet been solved satisfactorily (Abonyi et al 2003;Abonyi et al 2005;Liu et al 2008;Keogh and Kasetty 2003;Fuchs et al 2009Fuchs et al , 2010Fisch et al 2011;Kehagias et al 2005;Seghouane and Amari 2007). Kehagias et al (2005) presented a DP algorithm that uses Schwarz's Bayesian information criterion (BIC) as the segmentation order selection criterion (Seghouane and Amari 2007) to find the globally optimal segmentations for every value of segmentation order.…”