Evolutionarily optimized rough sets partitions

In this paper, an approach to optimally granulize rough set partition sizesusing evolutionary techniques is proposed. The evolutionary optimizationtechniques used are genetic algorithm particle swarm optimization and simulatedannealing. These evolutionary methods are compared to the hill climbingtechnique on an interstate conflict dataset. The results obtained from thisgranulization process are compared to the equal-width-bin andequal-frequency-bin partitioning techniques. The results obtained demonstratethat all of the proposed optimized methods produce higher forecasting accuraciesthan that of the two static methods and that genetic algorithm approach producethe highest accuracy. It is observed that the rules generated from the roughsets are linguistic and easy-to-interpret, but this does come at the expense ofloss of accuracy in the discretization process where the granularity of thevariables is decreased. ICIC International