Research on Data Classification Based on LNDS-SVDD

Due to the high dimension of industrial process data, there are aliasing areas when building hyper-spheres of different types of data by the SVDD algorithm, which will lead to inaccurate results in condition identification. A data classification method based on Local Neighbor Density Similarity Support Vector Data Description (LNDS-SVDD) is proposed to address this issue. Utilizing the local similar density in discriminating data similarity, the LNDS-SVDD can further judge the classification of samples distributed in the aliasing area using the density information. In this paper, a randomly generated synthetic data set and a real industrial process data set is used for simulating the proposed data classification algorithm. The results indicate that the classification accuracy is high, which is an effective condition identification method for the industrial process.