Evaluating Ranking Composition Methods for Multi-Objective Optimization of Knowledge Rules
HIS '08 Proceedings of the 2008 8th International Conference on Hybrid Intelligent Systems
A multi-objective GRASP for partial classification
Soft Computing - A Fusion of Foundations, Methodologies and Applications - Special Issue on Evolutionary and Metaheuristics based Data Mining (EMBDM); Guest Editors: José A. Gámez, María J. del Jesús, José M. Puerta
Rule induction for classification using multi-objective genetic programming
EMO'07 Proceedings of the 4th international conference on Evolutionary multi-criterion optimization
Evolutionary multi objective optimization for rule mining: a review
Artificial Intelligence Review
IEEE Transactions on Evolutionary Computation
Hi-index | 0.00 |
"If-Then" classification rules are one of many types of knowledge formats mined by data mining algorithms. These rules known as classifiers are evaluated based on objective and subjective metrics. Support, coverage, confidence, precision, recall, sensitivity and specificity are objective metrics which are used in measuring the accuracy of a classifier. Subjective measures like interest, surprise and comprehensibility are more user-oriented. Users prefer classifiers which are accurate, interesting and comprehensible. Hence mining classification rules with specific properties is considered as a multi objective optimization problem. Multi objective problems choose classifiers using optimization strategies which take the metric values as a vector or as a single functional value of the metrics taken together. Depending upon the values in the fitness vector the individuals may be dominators or pure dominators. In the current study an Extended Multi Objective Cultural Algorithm (EMOCA) framework is proposed for mining dominators and pure dominators taking coverage and confidence as optimization metrics and pure Pareto domination strategy to minimize the number of rules. The performance of dominators versus pure dominators in classifying unknown data instances and in producing a compact set of rules is compared using bench mark data set and the results reported.