基于改进距离孤立点检测预处理的LTSA算法

局部切空间排列算法(LTSA)是一种有效的流形学习方法,但该算法对孤立点的存在非常敏感.本文提出了一种快速有效的数据预处理方法-基于改进距离的孤立点检测方法来降低孤立点对LTSA算法的影响.该方法通过改进距离来度量样本点之间的距离,降低了样本点分布不均给孤立点检测算法带来的影响.实验表明,该数据预处理方法能有效地提高LTSA算法的鲁棒性,可以更好的挖掘数据集的本征特性,具有更好的数据可视化效果.

LTSA Algorithm Using Improved Distance Based Outliers Detection Preprocessing

The Local Tangent Space Alignment(LTSA) algorithm is an effective manifold learning method, but the algorithm is sensitive to outliers. A fast and effective data preprocessing procedure – improved distance based outliers detection method is presented to solve the problem in this paper. Using the improved distance to measure the distance between samples, the outlier detection method can reduce the negative influence brought by the imbalance distribution of samples. Experimental results demonstrate that the preprocessing procedure enables the LTSA algorithm to achieve more robust, and can discover the intrinsic structure of the dataset and can visualize the data better.