Abstract:

As an effective manifold-learning method, the local tangent space alignment （LTSA） algorithm is sensitive to outliers. In order to enhance the robustness of LTSA algorithm, an outlier detection method based on the improved distance is presented in this paper. In this method, the improved distance is used to measure the distance of the samples for the purpose of reducing the negative influence of the nonuniform distribution of the samples. Experimental results demonstrate that the proposed data preprocessing method can effectively improve the robustness of the LTSA algorithm and can discover the intrinsic characteristics of the dataset with better visualization effect.

Key words: data preprocessing, outlier detection, improved distance, manifold learning, local tangent space alignment