Resampling Methods on Classification Trees

• Main Authors: Wu, Han-Ming, 吳漢銘
• Other Authors: Yu-Shan Shih
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/92339876415811667275

碩士 === 國立中正大學 === 數理統計研究所 === 85 === Brieman's (1996) bagging and Freund and Schapire's (1996)boosting are recent resampling approaches to improvingpredictive accuracy of classification rules. Both methods combine multiple versions of unstable classifierssuch as classification trees to a composite classifier. In this paper, we study the applications of both techniquesto two tree- structured methods on a collection of datasets.The results show that, on average, both approaches can substantially improve predictive accuracy. But on some datasets consisting of influential observations, inferior results are obtained.A detection rule for influential points is then proposedon the basis of boosting algorithm. By removing influential observationsfrom the original learning sample, our results indicate thatbagging or boosting predictive accuracy.