Machine learning based aircraft arrival / departure registrations

Abstract

The aviation industry is vastly growing, as travelling by air is more common today than it ever was. However due too inefficiency and lack of communication of accurate flight information between airports, congestion and delays are occurring on a daily basis. While Collaborative Decision Making (CDM) is developed by Euro control to address this issue, the problem of transmitting accurate flight information near real time is not yet solved. Adecs Airinfra did a first attempt at automatic landing and departure registration by a fixed rule based algorithm to address this issue. However, this algorithm has limitations that cannot be solved with tweaking and tuning. In this work, we aim to create a replacement based on machine learning models. In this thesis we present the complete process, starting from raw real world data, turning this into
labelled data up to the point where we define a validation method and present the final results. We managed to create a machine learning landing / departure detection system with up to 99% precision and recall for arrivals, and for departures we managed to get a precision of 94% against 98% recall.