Abstract: With the help of a few labeled outlier samples，a supervised outlier detection based on Mahalanobis ellipsoidal learning machine（SODMELM） is proposed，which provides good description for the normal class as well as pushes the outlier samples away.This method inherits the advantage of Mahalanobis ellipsoidal learning machine（MELM） that considering the covariance matrix information of examples，i.e.，the distribution information of samples. Experiment results show that this method，on the average，can improve efficiency for outlier detection.

摘要： 针对有少量野点出现的情况，提出一种基于马氏椭球学习机的监督野点探测（supervised outlier detection based on Mahalanobis ellipsoidal learning machine，SODMELM）方法。这种方法通过一个超椭球对正常类进行较好的描述的同时，将野点排除在该椭球外面，继承了马氏椭球学习机（Mahalanobis Ellipsoidal Learning Machine，MELM）将样本点的协方差矩阵即样本点的分布信息考虑进去的优点。真实数据上的实验表明了所提的方法在一般意义上能提高野点探测的效率。