Rafael Bessa

Rafael Bessa

Data Scientist | ML Engineer

Publications

You can also find my articles on my Google Scholar profile.

Journal Articles

Predicting Porosity in Freeze Casting with Explainable Machine Learning

Published in ACS Omega, 2025

Freeze casting enables the fabrication of porous materials with tailored properties, but predicting porosity from process parameters is highly complex. In this work, we applied machine learning (CatBoost, Random Forest, XGBoost) to data from 252 studies across ceramics, polymers, and composites. CatBoost achieved the best performance (R² = 0.81). SHAP analysis identified solid loading as the main factor controlling porosity. Our results show that explainable ML can accelerate experimental design and support porosity optimization in freeze-cast materials.

Recommended citation: Rafael Gaspar Bessa de Oliveira, Jones Yudi, Edson Paulo da Silva, Guillermo Alvarez Bestard, Luiz Antonio Ribeiro Junior, and Alysson Martins Almeida Silva. ACS Omega 2025 10 (29), 32246-32256. DOI: 10.1021/acsomega.5c04133
Download Paper

Conference Papers

Model Aggregation for Medium-Term Prediction of Oil Reservoir Production

Published in Rio Oil and Gas Expo and Conference, 2022

In this study, we propose to forecast medium-term reservoir production without model calibration. We combine the data of multiple pre-existent models. A weight is then associated with each model regarding its relevance in forecasting production for the next time step. An optimization algorithm that minimizes a function based on the error of the projection of oil and water production in relation to the real reservoir data is considered for the computation of the weights.

Recommended citation: Bessa, Rafael & Limaverde Filho, José & Pimenta, Bráulio & Fortaleza, Eugenio & Munerato, Fernando & Soares de Melo Filho, Leonildes. (2022). Agregação de modelos para prever a produção de reservatórios de petróleo a médio prazo. Rio Oil and Gas Expo and Conference. 22. 36-37. 10.48072/2525-7579.rog.2022.036.
Download Paper