Clustering of Symbolically Described Events for Prediction of Numeric Attributes

This chapter describes a new conceptual clustering system capable of constructing classes for the prediction of a single numeric attribute. The clustering for single numeric attribute prediction (CSNAP) system clusters data described by a variety of symbolic attributes. The system trades off the accuracy of the predicted values against the clarity of the descriptions produced to produce classes which are both predictive and meaningful to a human observer. CSNAP has been used to develop classes for time based events and has demonstrated an ability to learn complex cyclic patterns.