论文部分内容阅读
马田系统是多变量数据挖掘中模式识别方法的新进展,变量间的复共线性会通过马氏距离影响马田系统变量筛选的效果和判别的准确率。为了克服复共线性对马田系统的负面影响,提出了基于岭估计新的测量尺度—岭马氏距离,通过变量敏感性和条件数绘制三条岭迹来确定岭参数,并设计了自适应多目标遗传算法进行基准空间优化,使得马田系统分类方法对病态数据具有更好的耐受性。通过案例验证了岭马氏距离可以很好的克服复共线性,并提高马田系统分类方法的判别准确率。
Martin system is a new development of pattern recognition in multivariate data mining. The complex collinearity between variables will affect the effectiveness of Martin’s system variable selection and the accuracy of discrimination through Mahalanobis distance. In order to overcome the adverse effect of complex collinearity on the Martin system, a new measurement scale based on ridge estimation is proposed, that is, Ling Mah-hsiung distance. Three ridge traces are drawn by variable sensitivity and conditional number to determine the ridge parameters. The target genetic algorithm is optimized for datum space, which makes the Martin classification more tolerant to pathological data. The case proved that the Ling Ma distance can be very good to overcome complex collinearity, and improve the classification accuracy of Martin field classification method.