population,it can handle multi-class learning in a natural way so that multiple classes can be learned simultaneously. The main process of the existing EA-based and stochastic search classification algorithms [36],is first generating rules randomly,and then improving the quality of the rules by using the training examples.So these algorithms adopt an up-bottom search mechanism,and may generate meaningless rules during the evolutionary process.But OCEC adopts a completely different search mechanism.For classification,if the obtained description is represented as rules,then each rule covers some examples,and the examples covered by the same rule have some similarities in attribute values.Based on this,OCEC first clusters the examples with similar attribute values so as to form organizations,and then guides the evolutionary process by using the differing significance of attributes.At the end of the evolutionary process,rules are extracted from organizations.Such a process can avoid generating meaningless rules. Therefore,OCEC adopts a bottom-up search mechanism. C.Organization ofpaper In the remainder of this paper,OCEC is described in Section II.Section III evaluates the effectiveness of OCEC by multiplexer problems.Section IV compares OCEC with the available algorithms on benchmarks,and applies OCEC to a practical case,the radar target recognition problem.The scalability of OCEC is studied in Section V.Finally,conclusions and some ideas for the future work are presented in the last section. >7 population, it can handle multi-class learning in a natural way so that multiple classes can be learned simultaneously. The main process of the existing EA-based and stochastic search classification algorithms [36], is first generating rules randomly, and then improving the quality of the rules by using the training examples. So these algorithms adopt an up-bottom search mechanism, and may generate meaningless rules during the evolutionary process. But OCEC adopts a completely different search mechanism. For classification, if the obtained description is represented as rules, then each rule covers some examples, and the examples covered by the same rule have some similarities in attribute values. Based on this, OCEC first clusters the examples with similar attribute values so as to form organizations, and then guides the evolutionary process by using the differing significance of attributes. At the end of the evolutionary process, rules are extracted from organizations. Such a process can avoid generating meaningless rules. Therefore, OCEC adopts a bottom-up search mechanism. C. Organization of paper In the remainder of this paper, OCEC is described in Section II. Section III evaluates the effectiveness of OCEC by multiplexer problems. Section IV compares OCEC with the available algorithms on benchmarks, and applies OCEC to a practical case, the radar target recognition problem. The scalability of OCEC is studied in Section V. Finally, conclusions and some ideas for the future work are presented in the last section