Document Type

Article

Publication Date

2025

Abstract

Voluntary safety reporting programs (VSRPs) can be essential in gaining insights into aviation safety and operations. However, analyzing safety reports can be time-consuming and tedious for subject matter experts (SMEs). AVIAN-S is a novel machine learning model that automatically labels aviation safety reporting data. This model utilizes an aviation-specific training set with over 70,000 samples of manually labeled aviation VSRP data. The model labels safety reports with a set of factors from a human factors taxonomy. Previous iterations of themodel utilized report rationales as the model input (Hinson, et al., 2023a). While utilizing rationales as input is helpful in the analysis of safety reports, it still includes the process of SMEs identifying rationales within the safety report. Including full report narratives as the model input can further expedite the report labeling process by eliminating the need to pull out specific rationales from thesafety report narrative. Building upon results presented at ISAP in 2023, the AVIAN-S model has been further developed to incorporate full report narrativesas the model input. This paper discusses results from the revised AVIAN-S model that uses full report narrative inputs and presents comparisons between the AVIAN-S model, a publicly available AI model, and SMEs.

Comments

Presented at the 23rd International Symposium on Aviation Psychology, May 27-30, 2025, Hosted by Oregon State University

Download


Share

COinS