Rough Sets Theory has its already large utility enhanced when a probabilistic approach is introduced in the applications with attributes presenting dominance relations (DRSA). Methods are here proposed to improve the quality of approximation, the number of reducts and the variety of decision rules in DRSA, based on the aggregation of classes of values of the decision attribute. The forms of aggregation proposed may be used in conjunction with other already known alternatives to DRSA. While these other strategies to elevate the value of the index of quality of approximation take into account only the cardinality of classes and border regions, the approach developed here employs also the distance between them. Two aggregation proposals are presented, one based on probability density and the other on probability of attaining extreme values.
Keywords: Rough Sets - Quality of Approximation - Dominance - Probability - Aggregation of Classes