Fair Causal Decision Tree

The decision tree algorithm is an effective machine learning technique, but it cannot uncover causal relationships within data. To overcome this limitation, the causal decision tree was proposed, which combines causal discovery with decision tree principles. This method is particularly effective at identifying causal relationships in complex datasets. Currently, causal decision tree algorithms are being applied in various fields, including biology and politics. However, existing causal decision tree algorithms do not account the fairness. To fill this gap, this paper proposes the Fair Causal Decision Tree (FCDT), constructed through a fair post-pruning process. The FCDT ensures the algorithm's fairness while maintaining the causality inherent in the causal decision tree. To preserve the algorithm's causality, this paper introduces the Causal Structure Change Constraint (CSCC), which guarantees that the causal decision tree retains its strong causal properties when handling unfair nodes. This innovative constraint ensures that any structural modifications during the fairness adjust-ment process almost not decrease the algorithm's causality. The experiments were conducted on multiple datasets, and the results demonstrate that the proposed algorithm enhances fairness by approximately 10% compared to traditional causal decision tree algorithms, with only about a 1 % decrease in recall rate.