Assessor Models for Predicting and Explaining Aleatoric Uncertainty in Classification

• Pedro B. S. Lima UFPE
• Ricardo B. C. Prudêncio UFPE
• Ana Carolina Lorena ITA
• Maria Gabriela Valeriano ITA

DOI: https://doi.org/10.5753/eniac.2024.245096

Resumo

Machine Learning (ML) models have been successfully adopted in several applications. However, despite their success, high levels of uncertainty can be observed depending on the problem being solved. Specifically, aleatoric uncertainty in classification is related to the inherent randomness associated with predictor and target attributes, usually observed for instances close to an area of class overlap. The literature on ML has developed different methods for quantifying uncertainty. The current paper addresses the complementary task of explaining uncertainty. For this, we proposed using assessors, which are meta-models trained to predict the performance of a given base model on a problem of interest. In our work, we adapted the concept of assessors to predict the aleatoric uncertainty of instances, measured by the average class entropy across a pool of diverse base models. Once built, eXplainable AI (XAI) techniques are adopted to extract explanations from the assessor. Experiments were performed in two case studies, using both a simulated and a real dataset for predicting the severity of COVID-19 patients. A pool of 12 base models was adopted to measure aleatoric uncertainty, and Random Forest was adopted as the assessor. The explanations extracted from the assessor were useful to identify specific features that cause high uncertainty in the classification problems.

Palavras-chave: Explainable AI, Machine Learning, Trustworthy Software and AI Systems

Referências

Antorán, J., Bhatt, U., Adel, T., Weller, A., and Hernández-Lobato, J. M. (2020). Getting a clue: A method for explaining uncertainty estimates. arXiv preprint arXiv:2006.06848.

Artelt, A., Visser, R., and Hammer, B. (2023). “i do not know! but why?”—local model-agnostic example-based explanations of reject. Neurocomputing, 558:126722.

Barek, M. A., Aziz, M. A., and Islam, M. S. (2020). Impact of age, sex, comorbidities and clinical symptoms on the severity of covid-19 cases: A meta-analysis with 55 studies and 10014 cases. Heliyon, 6(12).

Bhatt, U., Antorán, J., Zhang, Y., Liao, Q. V., Sattigeri, P., Fogliato, R., Melançon, G., Krishnan, R., Stanley, J., Tickoo, O., et al. (2021). Uncertainty as a form of transparency: Measuring, communicating, and using uncertainty. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, pages 401–413.

Burnell, R., Schellaert, W., Burden, J., Ullman, T. D., Martinez-Plumed, F., Tenenbaum, J. B., Rutar, D., Cheke, L. G., Sohl-Dickstein, J., Mitchell, M., et al. (2023). Rethink reporting of evaluation results in ai. Science, 380(6641):136–138.

Da Costa, D. C., Prudêncio, R., and Mota, A. (2023). Assessor models with a reject option for soccer result prediction. In 2023 International Conference on Machine Learning and Applications (ICMLA), pages 1200–1205.

Depeweg, S., Hernandez-Lobato, J.-M., Doshi-Velez, F., and Udluft, S. (2018). Decomposition of uncertainty in bayesian deep learning for efficient and risk-sensitive learning. In International conference on machine learning, pages 1184–1193. PMLR.

Hendrickx, K., Perini, L., Van der Plas, D., Meert, W., and Davis, J. (2024). Machine learning with a reject option: A survey. Machine Learning, 113(5):3073–3110.

Hernández-Orallo, J., Schellaert, W., and Martínez-Plumed, F. (2022). Training on the test set: Mapping the system-problem space in ai. In Proceedings of the AAAI conference on artificial intelligence, volume 36, pages 12256–12261.

Hüllermeier, E. and Waegeman, W. (2021). Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods. Machine learning, 110(3):457–506.

Liu, Z., Wang, Y., Vaidya, S., Ruehle, F., Halverson, J., Soljačić, M., Hou, T. Y., and Tegmark, M. (2024). Kan: Kolmogorov-arnold networks.

Lorena, A. C., Paiva, P. Y. A., and Prudêncio, R. B. C. (2023). Trusting my predictions: on the value of instance-level analysis. ACM Computing Surveys.

Molnar, C. (2022). Interpretable Machine Learning. 2 edition.

Peramo-Álvarez, F. P., López-Zúñiga, M. Á., and López-Ruz, M. Á. (2021). Medical sequels of covid-19. Medicina Clínica (English Edition), 157(8):388–394.

Prudencio, R. B., Lorena, A. C., Silva-Filho, T., and Drapal, P. (2024). Assessor models for explaining instance hardness in classification problems. In 2024 IEEE International Joint Conference on Neural Networks.

Valeriano, M. G., Kiffer, C. R., Higino, G., Zanão, P., Barbosa, D. A., Moreira, P. A., Santos, P. C. J., Grinbaum, R., and Lorena, A. C. (2022). Let the data speak: analysing data from multiple health centers of the são paulo metropolitan area for covid-19 clinical deterioration prediction. In 22nd IEEE CCGrid, pages 948–951. IEEE.

Watson, D., O’Hara, J., Tax, N., Mudd, R., and Guy, I. (2024). Explaining predictive uncertainty with information theoretic shapley values. Advances in Neural Information Processing Systems, 36.

Zhou, L., Martínez-Plumed, F., Hernández-Orallo, J., Ferri, C., and Schellaert, W. (2022). Reject before you run: Small assessors anticipate big language models. In EBeM IJCAI.