Forecasting Unstable Approaches with Boosting Frameworks and LSTM Networks​

This paper presents a machine learning algorithm trained to predict unstable approach events. Predictive modeling for unstable approaches (UA) forecasting needs a precursors analysis to determine the most important indicators (features) of aircraft instability. However, since the definition of aircraft instability is entirely dependent on the airline, these precursors might change according to the applied criteria. Most of the times, these precursors are related to the operation, ATC instructions, nearby weather conditions or even specific procedures for the selected airport or runway. We approached UA predictive analysis scenario from two different perspectives aligned with the same objective. On one hand, we performed the precursor analysis and binary classification using machine learning ensemble methodologies (boosting frameworks). On the other, we analyzed the FDM temporal series with Deep Learning techniques, using neural networks with Long Short Term Memory (LSTM) layers to binary classify if an unstable approach was about to happen and to detect unseen hazards or anomalies present in approach procedures.     «

This paper presents a machine learning algorithm trained to predict unstable approach events. Predictive modeling for unstable approaches (UA) forecasting needs a precursors analysis to determine the most important indicators (features) of aircraft instability. However, since the definition of aircraft instability is entirely dependent on the airline, these precursors might change according to the applied criteria. Most of the times, these precursors are related to the operation, ATC inst...     »