In practical use of a machine learning model, it is necessary to evaluate not only the performance metrics of the problem under consideration, but also the reliability of the predictions of this model. To improve reliability, uncertainty estimation methods and debiasing methods are used. This paper examines the relationship between these two components of reliability and proposes an effective way to combine these methods to achieve reliable predictions of the model.
Math-Net.Ru: https://www.mathnet.ru/eng/iipr628
Yu. Kuzmin, A. O. Shelmanov, I. V. Smirnov. Equal opportunity and uncertainty estimation in the text classification task // Artificial Intelligence and Decision Making, 2025, Issue 2, pp. 60–72.