Generalized K-Harmonic Means – Dynamic Weighting of Data in Unsupervised Learning
Bin Zhang · 2001
We propose a new class of center-based iterative clustering algorithms, K-Harmonic Means (KHMp), which is essentially insensitive to the initialization of the centers, demonstrated through many experiments. The insensitivity to initialization is attributed to a dynamic weighting function, which increases the importance of the data points that are far from any centers in the next iteration. The dependency of the K-Means' and EM's performance on the initialization of the centers has been a major problem. Many have tried to generate good initializations to solve the sensitivity problem. KHMp addresses the intrinsic problem by replacing the minimum distance from a data point to the centers, used in K-Means, by the Harmonic Averages of the distances from the data point to all centers. KHMp significantly improves the quality of clustering results comparing with both K-Means and EM. The KHMp algorithms have been implemented in both sequential and parallel languages and tested on hundreds of randomly generated datasets with different data distribution and clustering characteristics.