改进的直推式一类分类马氏椭球学习机

Improved transductive Mahalanobis ellipsoid learning machine for one class classification

针对采用马氏距离进行直推式学习的一类分类椭球学习机,在训练样本点较少而待分类样本点较多的情况中出现的处理较大规模数据集时间较长的问题,提出了一种改进的直推式马氏椭球学习机。采用样本协方差初始化策略构建初始化矩阵,在每次迭代中将距离当前超椭球中心最远的待分类样本点加入到学习机中形成新的训练集,进行训练直到得到最终的超椭球。通过对实际数据集进行实验验证,结果表明改进的算法在确保较高分类准确率的前提下,能有效地提高处理较大规模数据集的计算效率。

For the transductive mahalanobis ellipsoid learning machine for one class of classification,its running time is too long when dealing with large-scale data sets while the training sample points are less and samples to be classified are more.An improved transductive mahalanobis ellipsoidal learning machine is therefore proposed.Firstly,the initialization matrix is constructed by the sample covariance initialization strategy.In addition,the farthest sample points from the current hyper-ellipsoid centre are added into the learning machine at each iteration until get the final hyperellipsoid,and they form a new training set in order to retrain.Experimental results of real data sets show that the proposed algorithm can improve computing efficiency of the larger scale data sets on the basis of ensuring higher classification accuracy.