Mitigating bias in artificial intelligence: Fair data generation via causal models for transparent and explainable decision-making - ScienceDirect

Fig. 1. The figure depicts the flow of the proposed solution. (1) The first step would be to generate a causal model. (2) Apply the mitigation algorithm to the generated model. Finally, the mitigated model and the fair dataset are obtained. Fig. 2. DAGs are used to explain the mitigation rules. Green edges represent added relations, and red edges represent removed relations. Fig. 3. The figure illustrates two distributions, where the red represents the non-privileged group and the blue represents the privileged group. The X-axis reflects the level of education. Fig. 4. Compare the correlation between features before and after the mitigation process. Gray values indicate low correlation, while dark orange indicates high correlation. Table 1. Evaluation results for each model using performance metrics: Accuracy (A), True Positive Rate (TPR), False Positive Rate (FPR), False Negative Rate (FNR), and Predicted as Positive (PPP). The dataset used is the Original and Fair Data generated by m

saved by

• N . Soma Sekhar