Ex-ante Case Evaluation Study on Improving the Performance of CBR Classifier
In the CBR classifier, the classification accuracy and the reliability are two very important parameters that affect the generalization of the CBR system.Aiming at improving the accuracy of CBR classifier, and guarantee the reliability of the results, two ex-ante evaluation strategies based on the result confidence and a case revise method based on the feature weight distribution are proposed.Firstly, the evaluation methods are adopted to evaluate the confidence or trustworthiness of the reused results, which will divide the target dataset into trustworthy and untrustworthy sets;Secondly conduct the case revise on the untrustworthy set: utilize genetic algorithm to distribute the feature weight of the attributes, then use the attribute weight result to conduct a second round retrieve step on the untrustworthy set to fulfill the classification process.The experiment results showed that, the proposed methods could improve the accuracy of the CBR classifier, and could ensure the reliability of the classification result at the same time.
CBR classifier classification accuracy case evaluation feature weight distribution case revise
Zhao hui Yan ai-jun Wang pu
国际会议
昆明
英文
246-258
2014-05-01（万方平台首次上网日期，不代表论文的发表时间）