Ant Based Clustering of Time Series Discrete Data - A Rough Set Approach

This paper focuses on clustering of time series discrete data. In time series data, each instance represents a different time step and the attributes give values associated with that time. In the presented approach, we consider discrete data, i.e., the set of values appearing in a time series is finite. For ant-based clustering, we use the algorithm based on the versions proposed by J. Deneubourg, E. Ulmer and B. Faieta. As a similarity measure, the so-called consistency measure defined in terms of multistage decision transition systems is proposed. A decision on raising or dropping a given episode by the ant is made on the basis of a degree of consistency of that episode with the knowledge extracted from the neighboring episodes.