Attribution analysis method, device and electronic equipment based on decision mechanism

Abstract

Embodiments of the present disclosure disclose a decision mechanism-based attribution analysis method, device and electronic equipment. A specific implementation of the method includes: using feature data belonging to a target feature in the swap-in data set to train decision tree models of multiple depths to obtain multiple swap-in candidate decision tree models; using feature data belonging to the target feature in the swap-out data set to train decision tree models of multiple depths to obtain multiple swap-out candidate decision tree models; selecting swap-in decision tree models and swap-out decision tree models of the same depth according to index values of each candidate decision tree model and the original decision model on a preset constraint index; and based on decision rules generated by the swap-in and swap-out decision tree models, analyzing differences in the decision mechanism of the original decision model for the swap-in and swap-out data sets. The implementation is related to model bottom layer analysis technology, and by analyzing the decision rules of the decision tree, the bottom layer decision logic causing the swap-in and swap-out is revealed.

Images