Go-Around Prediction in Non-Stabilized Approach Scenarios Through a Regression Machine-Learning Model Trained from Pilots’ Expertise

Abstract

To face the pilots’ shortage that could hit the aeronautical world in the future, single-pilot operations are envisaged as a solution. Even if the single pilot will have to assume the tasks that are today done by two pilots in the cockpits, the safety level of the flight should remain the same compared to standard operations. The challenge will be significant during high workload phases when complex decisions are to be made. We focused our work on the final approach phase during which the decision to perform a go-around or to land must be taken. This decision reveals to be particularly difficult as studies show that 97% of the time, the choice to land is made whereas a go-around was needed. In this paper, we propose a machine learning regression algorithm based on expert users classification to predict the need for a go-around during the final approach. An average precision of 0.96, a sensitivity of 0.84 and an F1-score of 0.88 shows a promising behaviour for the GA automatic identification task.

This work has received funding from the Clean Sky 2 Joint Undertaking (JU) under grant agreement No 831884. The DGX A100 used for this research has been funded by the European Union within the operating Program ERDF of the Valencian Community 2014–2020 with the grant number IDIFEDER/2020/030.