Weighted Cross-Product Constraint Transformation to Optimize Spatial Structure of Data
Siqing Wang, Deqi Li, Xin Zhang, Shutao Zhang · 2021
The research of data spatial structure optimization is significant in data mining. Many methods optimize the space based on a specific criterion, which in a certain extent destroys the original spatial distribution of data with uncertain optimization effects. Based on the heterogeneous distribution spatial structure of the data and the importance of the data attributes to distinguish categories, we proposed the weighted cross-product constraint transformation (WCCT) method to improve the accuracy of data analysis. This method retains the spatial distribution of the original data and maps the data to a limited high-dimensional space with constraints, which increases the proximity between the same data category and the discrimination between different data categories. Moreover, we established a statistical evaluation system of spatial structure to prove the effectiveness of this method in optimizing the spatial structure of data attributes. Experimental analysis on real data demonstrates that the proposed method obtains excellent classification performance with different machine learning classifiers.