A Case Study of Accurate and Fair Classification

As artificial intelligence and human life become increasingly inseparable, the legal or ethical issues faced by artificial intelligence systems in autonomous decision-making are also increasing. During the training process, the algorithms may be influenced by human biases such as gender, race, and other factors, leading to discrimination and affecting fairness. To establish secure intelligent systems, fair machine learning has become a popular research direction. This work demonstrates the existing definitions of fairness and designs experiments to show that combining clustering algorithms into the data handling process can effectively improve the classification accuracy and fairness on bank loan dataset. In the case study, K-means clustering, hierarchical clustering and Gaussian Mixture Model are used, proving that the clustering algorithms can significantly improve the accuracy of the model and ensure the relative fairness of the classification results.