SVM and reduction-based two algorithms for examining and eliminating mistakes in inconsistent examples

• Published in: Proceedings of 2004 International Conference on Machine Learning and Cybernetics (IEEE Cat. No.04EX826) Vol. 4; pp. 2189 - 2192 vol.4
• Main Authors: Hong-Hai Feng, Ming-Yi Liao, Guo-Shun Chen, Bing-Ru Yang, Yu-Mei Chen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2004
• Subjects: Agricultural engineering; Chemical engineering; Chemical technology; Data engineering; Economic forecasting; Electronic mail; Finance; Optical fiber devices; Support vector machines; Testing
• ISBN: 0780384032; 9780780384033
• DOI: 10.1109/ICMLC.2004.1382161

Currently, misleading findings during data mining are mostly caused by the dirty data including that caused by mistakes of inconsistent examples. A reduction-based algorithm for helping us to check the mistakes hidden in inconsistent examples is presented, i.e., firstly, through reduction to eliminate the attributes which have no relation to the inconsistencies and leave behind the ones which have some relations to the inconsistencies; secondly, to examine whether there are some mistakes of the core attribute values of the inconsistent examples in the reduction; finally, to examine the non-core attribute values of the inconsistent examples. When the amount of the inconsistent examples is large, the algorithm is especially necessary. In addition, a SVM based algorithm is presented for eliminating the inconsistent examples in which there are mistakes. Using SVM, the consistent examples are trained first, the inconsistent examples are tested, the inconsistent examples that fall into the test class are left behind, and the inconsistent examples that do not belong to the test class are eliminated.