A data-driven model for safety risk identification from flight data analysis

Abstract

Most aviation accidents take place in the final phase of a flight. One possible accident is the runway overrun - the fact that an aircraft leaves the runway unexpectedly on landing. Even though such accidents are well documented and studied in the aviation industry, this paper aims at identifying less direct links between data recorded by planes and the risk of runway overrun, or linked events. Indeed, a better understanding of these events using available flight data helps to reduce their number. Nonetheless, such analysis is not straightforward given the massive volume of data collected during the flights. For that purpose, we propose a data-driven approach with the use of data analysis methods and machine learning tools. After a quick correlation analysis, a boosted tree classifier was trained to classify flights as safe or at risk. The classifications were accurate enough to extract contributing factors, and a more extensive analysis was conducted on multiple airports. That analysis revealed the importance of particular factors, leading to new insights about potential approaches to aviation safety.

Keywords

• Air transportation
• Decision-support systems
• Risk prediction
• Artificial intelligence,