Explainable Boosting Machine: A Contemporary Glass-Box Strategy for the Assessment of Wind Shear Severity in the Runway Vicinity Based on the Doppler Light Detection and Ranging Data

Abstract

Pilots commonly undergo training to effectively manage instances of wind shear (WS) during both the landing and takeoff stages. Nevertheless, in exceptional circumstances, there may be instances of severe wind shear (SWS) surpassing a magnitude of 30 knots, leading to adverse effects on the operation of taking off and landing aircraft. This phenomenon can lead to the execution of aborted landing maneuvers and deviations from the intended glide path. This study utilized the explainable boosting machine (EBM), an advanced machine learning (ML) model known for its transparency, to predict the severity of WS occurrences and analyze the underlying factors. The dataset consisted of 21,392 data points from 2018 to 2022 acquired from two Doppler light detection and ranging (LiDAR) systems installed at Hong Kong International Airport (HKIA). Initially, the Doppler LiDAR data received data treatment in order to address the issue of data imbalance. Subsequently, utilizing the processed data, the hyperparameters of EBM were optimized using the Bayesian optimization technique. The EBM model underwent subsequent training and evaluation, wherein its performance metrics were computed and compared with those of an alternative glass-box model including decision tree (DT) and counterpart black-box models, namely, random forest (RF) and extreme gradient boosting (XGBoost). The EBM model trained on synthetic minority oversampling technique (SMOTE)-treated data demonstrated superior performance in comparison with the alternative models, as indicated by its higher geometric mean (0.77), balanced accuracy (0.78), and Matthews’ correlation coefficient (0.169). Furthermore, the EBM exhibited enhanced predictive performance and facilitated a comprehensive analysis of individual and pairwise factor interactions in the prediction of WS severity. This enabled the assessment of the factors that contributed to the instances of SWS in the proximity of airport runways.