Using Machine Learning to Improve Regulatory Review of Flight Waivers and Exemptions
Carlos E. Martinez, David S. Day · 2018
Through development of a MITRE-funded prototype tool, we have demonstrated the feasibility of applying machine learning and Natural Language Processing (NLP) to help aviation regulatory agencies more efficiently, consistently, and effectively analyze requests to operate unmanned aircraft consistent with a complex set of operating rules and regulations. The prototype, called ARDeST (Application Review Decision Support Tool), uses NLP models derived from machine learning to aid in reviewing large numbers of applications requesting waivers or exemptions from existing aviation regulations. The prototype attempts to predict the likely decision outcome for each pending application, thereby helping agencies to triage incoming requests. It then provides decision support to the individual regulatory reviewer by identifying and highlighting key decision features that relate to issues that must be addressed for regulatory approval. All of the automated analyses are associated with confidence metrics derived from the underlying statistical machine learning algorithms employed, which can help decision makers in their prioritization and review process. By highlighting decisional factors and the sentences or paragraphs in which they are addressed, reviewers are able to focus principally on the most important material. This tool increases efficiency, reduces the cognitive load on the reviewer, and enables rapid decision making.