Semi-supervised Evolutionary Distance Metric Learning for Clustering

Exising method for supervised clustering called Evolutionary Distance Metric Learning (EDML) has never been compared to other clustering method. This work conducted experiments to compare EDML with other semisupervised clusterings, such as COP-Kmeans and other DML methods. The result empirically confirms that EDML gives better clustering structure than the candidate clustering methods-i.e. K-means, COP-Kmeans, and MPC-Kmeans. Also, we justify the effect of the number of constraints, effect of smoothing, and the feasibility to evaluate EDML in various criteria. Therefore, EDML is assured that it has potential to improve clustering quality and is capable of using various clustering indices.