On the Statistics and Predictability of Go-Arounds

TL;DR

This study analyzes operational and weather factors at San Francisco Airport that precede go-arounds, developing a machine learning-based alert system to predict such maneuvers and improve air traffic management.

Contribution

It provides an empirical analysis of factors leading to go-arounds and introduces a preliminary machine learning alert system for prediction.

Findings

Operational factors like inbound aircraft count fluctuate before go-arounds

Weather conditions influence go-around likelihood

The alert system shows promise for real-time prediction

Abstract

This paper takes an empirical approach to identify operational factors at busy airports that may predate go-around maneuvers. Using four years of data from San Francisco International Airport, we begin our investigation with a statistical approach to investigate which features of airborne, ground operations (e.g., number of inbound aircraft, number of aircraft taxiing from gate, etc.) or weather are most likely to fluctuate, relative to nominal operations, in the minutes immediately preceding a missed approach. We analyze these findings both in terms of their implication on current airport operations and discuss how the antecedent factors may affect NextGen. Finally, as a means to assist air traffic controllers, we draw upon techniques from the machine learning community to develop a preliminary alert system for go-around prediction.