CPC G06F 21/6245 (2013.01) [H04L 9/085 (2013.01); H04L 2209/46 (2013.01)] | 20 Claims |
1. A computer-implemented method for privacy protection-based multicollinearity detection, comprising:
performing, by one of at least two member devices, data alignment on respective local feature data with other one or more member devices of the at least two member devices, to construct a joint feature matrix, wherein each of the at least two member devices holds the respective local feature data;
performing, by the one of at least two member devices, privacy protection-based multi-party matrix multiplication computation with the other one or more member devices, to compute a product matrix of a transposed matrix of the joint feature matrix and the joint feature matrix, wherein each member device holds a submatrix of the product matrix;
determining, by the one of at least two member devices, an inverse matrix of the product matrix with the other one or more member devices based on respective submatrices of the product matrix, wherein each member device holds a submatrix of the inverse matrix;
determining, by the one of at least two member devices, a variance inflation factor of each attribute feature with the other one or more member devices using respective submatrices of the inverse matrix and the respective local feature data, wherein each member device holds fragment data of the variance inflation factor of each attribute feature; and
determining, by the one of at least two member devices, multicollinearity with the other one or more member devices based on respectively held fragment data of the variance inflation factor of each attribute feature.
|