Distribution-based aggregation for relational learning with identifier attributes