异常值检验方法的比较分析

Comparative Analysis of Outlier Test Methods

针对格鲁布斯检验法、t检验法检测异常值的局限性,使用格鲁布斯检验法、t检验法、跳跃度检验法对异常值的个数分七种情况进行模拟研究,并对工效学用户测评数据的异常值检验结果进行比较分析,研究结果表明,跳跃度检验法识别异常值的准确率更高,并且在一定数量的异常值下避免了Masking效应(判异为正)和Swamping效应(判正为异)。当数据差异较小时,格鲁布斯检验法和t检验法检测异常值的效果较好,当数据差异较大时,跳跃度检验法检测异常值的效果更好。

Aiming at the l imitations of Grubbs’ test method and T-test method, These are studied that a simulation study on the number of outl iers in seven cases, by using the Grubbs, test method, T-test meth-od and jump test method, and a comparative analysis of outlier results about the data of ergonomics evalua-tion by user test. It can be proved that Jump test method, the accuracy is more accurate in identifying out-lier, and the method can avoid the Masking effect （outl iers are judged to be normal values） and Swamping effect （normal values are judged to be outliers） in a certain amount of outl iers. Grubbs’ test method and T-test method have a better effect on detecting outliers, when the difference of data is small, and the jump test method has a better effect on detecting outl iers are also confirmed, when the difference of data is large.