Clustering is an important technique in data mining and machine learning in which underlying and meaningful groups of data are discovered. One of the paramount issues in clustering process is to discover the natural groups in the data set. A method based on the minimum message length (MML) has been developed to determine the optimum number of clusters (or mixture model size) in a power quality data set from an actual harmonic monitoring system in a distribution system in Australia. Once the optimum number of clusters is determined, a supervised learning algorithm, C5.0, is used to uncover the fundamental defining factors that differentiate the various clusters from each other. This allows for explanatory rules of each cluster in the harmonic data to be defined. These rules can then be utilised to predict which cluster any new observed data may best be described.
Available at: http://works.bepress.com/dstirling/36/