Missed Approach, a Safety-Critical Go-Around Procedure in Aviation: Prediction Based on Machine Learning-Ensemble Imbalance Learning

Abstract

The final approach phase of an aircraft accounts for nearly half of all aviation incidents worldwide due to low-level wind shear, heavy downpours, runway excursions, and unsteady approaches. Adopting the missed approach (MAP) procedures may prevent a risky landing, which is usually executed in those situations, but it is safety-critical and a rare occurrence. This study employed machine learning-ensemble imbalance learning to predict MAPs under low-level wind shear conditions based on environmental and situational parameters. The models were developed using the 2017–2021 Hong Kong International Airport (HKIA) Pilot Reports (PIREPs). Initially, imbalance data were applied to machine learning models such as the random forest (RF), light gradient boosting machine (LGBM), and extreme gradient boosting (XGBoost), but these were unable to accurately predict the occurrence of MAPs. Then, these models were used as base estimators for ensemble imbalance learning methods, including the self-paced ensemble (SPE) framework, the balance cascade model, and the easy ensemble model. The SPE framework utilizing XGboost as the base estimator performed better than other frameworks in terms of recall, F1-score, balanced accuracy, and geometric mean. Afterwards, SHAP was utilized to interpret the SPE framework with XGboost as the base estimator. Results showed that low-level wind shear magnitude, runway orientation, and vertical location of low-level wind shear contributed most to MAPs. Runways 07C and 07R had the most MAPs. Most MAPs were initiated when low-level wind shear was within 500 feet of the ground. Strong tailwind triggered MAPs more than headwind. For aviation safety researchers and airport authorities, the framework proposed in this study is a valuable tool.

An Imbalanced Data Classification Method Based on Hybrid Resampling and Fine Cost Sensitive Support Vector Machine

“…Based on these, multiple XG-Boost classifiers were trained, and then the outputs were integrated by using a simple voting method to obtain the final classifier. Khattak et al [24] proposed an improved Balance Cascade algorithm that integrates Bootstrap and XG-Boost, in which, the positive samples are sampled using Bootstrap and the negative samples are sampled using the Balance Cascade algorithm, while XG-Boost is used as the base classifier. Zhao et al [25] proposed an imbalanced data classification method based on under-sampling and an improved SVM.…”