Improving the performance of inductive learning classifiers through the presentation order of the training patterns

摘要: Abstract Although the development of new supervised learning algorithms for machine techniques are mostly oriented to improve predictive power or classification accuracy, capacity understand how process is carried out great interest many applications in business and industry. Inductive algorithms, like Rules family, induce semantically interpretable rules form if-then rules. effectiveness family has been studied thoroughly improved versions constantly developed, one important drawback effect presentation order training patterns which not depth previously. In this paper issue addressed, first by studying empirically random orders number generalization resulting classifier. Then a method examples proposed combines clustering stage with density measure developed specifically problem. The results using benchmark datasets real application wood defect show method. Also, since employed as preprocessing stage, simplicity affected but instead it enables generation fewer more accurate rules, can have direct impact performance usefulness an expert system context.