Systematic evaluation of machine learning and explainable AI for NOx emission prediction in gas turbine systems
Accurate prediction of nitrogen oxide (NOx) emissions is essential for improving environmental sustainability, regulatory compliance, and operational efficiency in gas turbine systems. However, the complex nonlinear interactions among environmental and operational variables make reliable emission forecasting challenging using conventional modeling approaches. This study presents a comprehensive machine learning and explainable artificial intelligence (XAI) framework for NOx emission prediction using the Gas Turbine CO and NOx Emission dataset from the UCI Machine Learning Repository. Ten regression algorithms, including linear, tree-based, ensemble, and gradient-boosting models, were systematically evaluated and optimized using Grid Search, Random Search, and Bayesian Optimization. Model performance was assessed using multiple regression metrics, including R², MAE, MSE, RMSE, MAPE, EVS, and MedAE. Among all evaluated approaches, Bayesian-optimized LightGBM achieved the best predictive performance with an R² score of 0.8928 and an RMSE of 3.7694. To enhance model transparency, multiple complementary XAI techniques, including SHAP, Permutation Feature Importance, Partial Dependence Plots, Individual Conditional Expectation, LIME, surrogate modeling, and residual analysis, were employed. The explainability results consistently identified ambient temperature, turbine inlet temperature, gas turbine exhaust pressure, and turbine after temperature as the dominant factors influencing NOx emissions. The proposed framework demonstrates that high predictive accuracy and interpretability can be achieved simultaneously, providing a promising and transparent approach for NOx emission prediction in gas turbine systems.
Authors
- Sandhyalaxmi G Navada (ORCID: https://orcid.org/0000-0002-2305-9297)
- Omkar Prabhu (ORCID: https://orcid.org/0009-0002-0990-7365)
Institutions
- Manipal Academy of Higher Education (IN)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1038/s41598-026-80856-9
- Primary Topic
- Explainable Artificial Intelligence (XAI)
- Type
- article
- Field-Weighted Citation Impact
- 0.00