Bottom-up Pittsburgh Approach for Discovery of Classification Rules

This paper presents bottom-up Pittsburgh approach for discovery of classification rules. Population initialization makes use of entropy as the attribute significance measure and contains variable sized organizations. Each organization contains a set of IF-THEN rules. As bottom-up approach is employed, so traditional operators are not feasible and efficient to use. Therefore, four evolutionary operators are devised for realizing the evolutionary operations performed on organizations. Bottom-up Pittsburgh approach gives best set of rule having good accuracy. In experiments, the effectiveness of the proposed algorithm is evaluated by comparing the results of bottom-up Pittsburgh with and without entropy to the top-down Michigan approach with and without entropy on 10 datasets from the UCI and KEEL repository. All results show that bottom-up Pittsburgh approach achieves a higher predictive accuracy and is more consistent.