Fairness measures for educational datasets

This paper aims to provide a practical and reproducible method for measuring the fairness of classification algorithms. We highlight problems that may arise both with datasets and predictions. Through a simple yet clear approach, we show how to consistently measure fairness based on statistical non-discrimination criteria. As education plays a crucial role in human life, aligning Artificial Intelligence with ethical values becomes particularly important in this context. For this reason, we choose a practical example representative of an educational environment, namely a public dataset with typical students’ features.We present a straightforward, understandable, and comprehensive fairness measure that takes into account Independence, Separation, and Sufficiency criteria focusing on a method to calculate the Calibration fairness measure.