Balancing Fairness and Interpretability in Clustering: Introducing FairParTree

The revolution involving Machine Learning and Artificial Intelligence has transformed data analytics, making algorithms play important roles in decision-making processes across various domains, even in sensitive scenarios. Indeed, traditional clustering algorithms often lack interpretability and exhibit biases, leading to discriminatory practices and opaque decision-making. We introduce FairParTree, a novel fair and interpretable clustering algorithm built upon the ParTree clustering algorithm. FairParTree integrates fairness constraints directly into the clustering process, making sure that the resulting clusters do not disproportionately disadvantage any particular group. The algorithm employs three fairness definitions: demographic fairness, individual fairness, and group fairness. By leveraging the structure of decision trees, FairParTree also enhances the interpretability of clustering results, providing clear and understandable explanations for cluster assignments. We evaluate FairParTree's performance against state-of-the-art competitors. Through extensive experiments, we show that it maintains strong performances in w.r.t. fairness, explainability, and clustering quality across different dataset sizes, asserting its value as a fair, explainable, and efficient clustering algorithm.