Mean-tracking as an Efficient Preprocessor for the K-means Clustering Algorithm

R. Mitchell (UK)


Clustering Algorithms, Load Forecasting


Many algorithms have been developed for clustering data, of which the K-means algorithm is one of the most popular. There are, however, problems with the algorithm, for instance, the number of clusters, K, needs to be known; and the result is affected by the initial positions of the cluster search. This paper describes a novel hybrid algorithm designed to overcome these disadvantages, in which the mean-tracking algorithm is used as a pre-processing stage, to find the initial values for the k-means algorithm. The success of the algorithm is demonstrated, improvements to the mean-tracking method are given, and it is shown to be faster than k-means. The algorithm, although general, is applied successfully to electricity load forecasting data, enabling demand to be predicted accurately.

Important Links:

Go Back