A data preprocessing method based on a relevance vector machine and a radial basis function network

The presence of outliers in the dataset and a too concentrated data distribution affects the performance of the data-driven model. In this paper, the relevance vector machine (RVM) and the radial basis function (RBF) network are used for data preprocessing. First, by gradually reducing the width of the kernel function, the RVM regression model gradually approaches the input sample to be tested, and the outlier is identified by the 3σ criterion; second, the attribute value of the abnormal probability is introduced to reduce the interference of the abnormal value upon determining the normal value; Finally, the RBF network is used to fill in the deleted outliers, and a certain number of virtual samples are generated in the small sample interval. The comparative experiment shows that this method can effectively improve data quality.