Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Human Factor Identification in Aviation Accidents Using Contextual Word Embeddings

July Bias Macêdo1, Plínio M. S. Ramos2,a, Caio Souto Maior2,b, Márcio J. C. Moura2,c and Isis Didier Lins2,d

1Center for Risk Analysis, Reliability Engineering and Environmental Modeling (CEERMA). Department, Department of Industrial Engineering, Federal University of Pernambuco (UFPE), Brazil.

2CEERMA. Department, Department of Industrial Engineering, UFPE, Brazil.


Human error is a leading cause of aviation accidents and can result in significant loss of life. To support decision-making, the aviation industry collects extensive data, including written accident investigation reports that contain valuable information for risk analysis and accident management. Natural Language Processing (NLP) can assist experts in processing and analyzing these reports, enabling effective risk management and the proposal of preventive measures. This paper proposes a novel methodology for identifying human factors leading to aviation accidents using topic modeling based on contextual word-vector representations extracted from pre-trained Bidirectional Encoder Representation from Transformers (BERT). The proposed approach differs from previous studies identified in a systematic literature review. This methodology can provide useful insights for proposing preventive measures and training plans to reduce the risk of human error.

Keywords: Aviation accidents, Human factors, Topic modelling, Aviation safety, Natural language processing.

Download PDF