Nearest neighbours in least-squares data imputation algorithms with different missing patterns