Shape Identification in Temporal Data Sets

Abstract

Shapes are a concise way to describe temporal variable behaviors. Some commonly used shapes are spikes, sinks, rises, and drops. A spike describes a set of variable values that rapidly increase, then immediately rapidly decrease. The variable may be the value of a stock or a person’s blood sugar levels. Shapes are abstract. Details such as the height of spike or its rate increase, are lost in the abstraction. These hidden details make it difficult to define shapes and compare one to another. For example, what attributes of a spike determine its “spikiness”? The ability to define and compare shapes is important because it allows shapes to be identified and ranked, according to an attribute of interest. Work has been done in the area of shape identification through pattern matching and other data mining techniques, but ideas combining the identification and comparison of shapes have received less attention. This paper fills the gap by presenting a set of shapes and the attributes by which they can identified, compared, and ranked. Neither the set of shapes, nor their attributes presented in this paper are exhaustive, but it provides an example of how a shape’s attributes can be used for identification and comparison. The intention of this paper is not to replace any particular mathematical method of identifying a particular behavior, but to provide a toolset for knowledge discovery and an intuitive method of data mining for novices. Spikes, sinks, rises, drops, lines, plateaus, valleys, and gaps are the shapes presented in this paper. Several attributes for each shape are defined. These attributes will be the basis for constructing definitions that allow the shapes to be identified and ranked. The second contribution is an information visualization tool, TimeSearcher: Shape Search Edition (SSE), which allows users to explore data sets using the identification and ranking ideas in this paper.