Nitrogen oxide (NO \(_x\) ) emissions from aviation are a critical environmental concern, often exacerbated by aircraft engine performance degradation. However, computational models capable of detecting such deterioration based specifically on the NO \(_x\) Landing and Take-Off (LTO) total mass are currently lacking. This study proposes a data-driven anomaly detection framework utilizing the July 2024 release of the ICAO Aircraft Engine Emissions Databank (EEDB), encompassing 730 unique engine configurations. The methodology involved the selection of 13 primary input parameters, Min-Max normalization within the [0, 1] range, and a training strategy based on an 80%–20% data partition and 5-fold cross-validation to ensure model generalizability. Among the evaluated architectures–Multi-Layer Perceptron (MLP), Support Vector Machines (SVM), and Gaussian Process Regression (GPR)–the Rational Quadratic GPR (RQGPR) model demonstrated superior predictive fidelity. The RQGPR model achieved an average correlation coefficient (R) of 0.99640, a coefficient of determination ( \(R^{2}\) ) of 0.99272, and a minimized mean squared error (MSE) of 0.00037. Leveraging these high-precision outputs, 99% prediction intervals were established to serve as a statistical baseline for nominal engine operation. Emissions falling beyond these defined boundaries are flagged as statistical anomalies, providing a robust proxy for performance degradation or mechanical discrepancies with respect to certified emission standards. This non-intrusive approach offers airline operators and regulatory authorities a high-precision diagnostic tool for optimizing maintenance schedules and ensuring strict adherence to environmental regulations.
