Similarity-based classification of microdata