Natural Language Processing and Data-Driven Methods for Aviation Safety and Resilience: from Extant Knowledge to Potential Precursors

Demand upon the future Air Traffic Management (ATM) system will possibly grow to exceed available system capacity, pushing forward the need for automation and digitisation to maintain safety while increasing efficiency. This work focuses on a manifestation of ATM safety, the loss of separation (LoS), and its analysis via Natural Language Processing (NLP) and Data-Driven Methods (DDMs), able to extract meaningful and actionable information from the LoS-related data. These data are, primarily, safety reports and ATM-system data (e.g, flights information, radar tracks, and Air Traffic Control events). Current research in this field mainly exploits NLP to categorise the reports and DDMs to predict safety events. The limitation of current NLP-based approaches is that the considered categories need to be manually annotated by experts and general taxonomies are seldom exploited. At the same time, current DDMs are rarely able to support safety practitioners in the process of investigation of an incident after it happened. To fill these gaps, the authors propose to (i) perform Exploratory Data Analysis on safety reports combining state-of-the-art techniques like topic modelling and clustering, then to (ii) develop an algorithm able to extract the recent Toolkit for ATM Occurrence Investigation (TOKAI) taxonomy factors from the free-text safety reports based on Syntactic Analysis, and finally to (iii) develop a DDM able to automatically assess if the Pilots or the Air Traffic Controller (ATCo) or both contributed to the incident, almost immediately after the LoS. The results on LoSs reported in the public database of the Comisión de Estudio y Análisis de Notificaciones de Incidentes de Tránsito Aéreo (CEANITA) support the authors' proposal.