Model-based Classification of Data with Time Series-valued Attributes

Similarity search and data mining on time series databases has recently attracted much attention. In this paper, we represent a data object by several time series-valued attributes. Although this kind of object representation is very natural and straightforward in many applications, there has not been much research on data mining methods for objects of this special type. In this paper, we propose a novel model-based classifier exploiting the benefits of representing objects by several time series. Classification decisions are based on class-specific interaction patterns among the time series of an object. Experimental results on benchmark data and real-world medical data demonstrate the performance of our classifier.