Font Size: 

To evaluate the performance of clustering algorithms is challenging because typically the true classes are unknown. In this paper we propose a new cluster validity method that combines internal and relative criteria and employs Machine Learning algorithms to produce a relative validity ranking of partitions obtained from different clustering algorithms. Compared to other methods, the proposed approach considers the features' structure explicitly, can handle high-dimensional data, and can be applied to various clustering algorithms. The method has been tested on a simulated benchmark dataset, demonstrating its ability to rank correctly 11 classical clustering algorithms.