Statistical methods for modeling the risk of runway excursions

The goals of this paper are to: (i) enhance the manner in which fatal airport runway excursions are modeled and quantified and (ii) explore a means to mitigate their occurrence and severity. While other research in predicting runway excursions has focused on the type of excursion, this work focuses on predicting if the excursion will generate fatalities. As the adverse effects of fatalities can be extreme in nature, there exists a need to be able to: (i) understand the root causes of fatal excursions, (ii) predict the likelihood of fatal excursions, and (iii) measure the efficacy of risk management strategies employed to prevent them. This work summarizes and applies techniques of data analysis for runway excursions, a significant problem in air travel safety which can lead to fatalities. The techniques deployed in this work to model excursions include logistic regression and Bayesian logistic regression, each of which have strengths and weaknesses in terms of descriptive (e.g. highlighting factors that impact fatalities) and prescriptive (e.g. predicting fatalities under particular operating conditions) domains. An innovative use of the results of this data analysis is in enhancing the likelihood assessment of the traditional risk matrix, which combines (often arbitrary) assessments of likelihood and consequence for particular risk scenarios. Several real-world excursion response options aimed at reducing fatalities through improvements to aviation facilities and processes are compared on the basis of impact, cost, and feasibility.