Research on transform of outliers based on density

This paper proposed a data transform method based on error-adjusted density of micro-datasets, so as to distinguish the characteristics of outliers efficiently and improve the accuracy of prediction models. It divided the large multi-dimensional data sets into many grid cells, and in each cell assigned each data point to its closest micro-dataset using a nearest neighbor algorithm, data points were represented by calculating the error-adjusted density estimation in each micro-dataset. Thereby, the processed data could embody the information of the area which they belonged to and show the data variation characteristics rightly.