Band depth based initialization of K-means for functional data clustering

Abstract The k -Means algorithm is one of the most popular choices for clustering data but well-known to be sensitive initialization process. There a substantial number methods that aim at finding optimal initial seeds -Means, though none them universally valid. This paper presents an extension longitudinal such methods, BRIk algorithm, relies on set centroids derived from bootstrap replicates and use versatile Modified Band Depth. In our approach we improve method by adding step where fit appropriate B-splines observations resampling process allows computational feasibility handling issues as noise or missing data. We have two techniques providing suitable seeds, each stressing respectively multivariate functional nature Our results with simulated real sets indicate F unctional Data A pproach BRIK (FABRIk) D ata E xtension (FDEBRIk) are more effective than previous proposals initialize in terms recovery.