Human Factors in Aviation Safety: NLP on Pilot Reports

Aviation safety continues to be heavily influenced by human factors, contributing to more than 70% of documented incidents. This study proposes a Natural Language Processing (NLP)-based framework designed to analyze pilot reports from the Aviation Safety Reporting System (ASRS) to uncover key human factors such as lapses in situational awareness, failures in communication, and cognitive overload. Three hybrid deep learning models were implemented and evaluated: CNN + BiLSTM + BiGRU + Attention, CNN + BiLSTM + Transformer + Multi-Head Self-Attention, and CNN + BiGRU + Transformer + Multi-Head Self-Attention. Semantic representations were generated using pre-trained Word2Vec embeddings, ^whil^e the models were trained on a balanced subset of ASRS narratives labeled for the presence of gsituational awareness” as a binary classification problem. To enhance model generalization, data augmentation methods such as synonym substitution and random word rearrangement were applied. Among the proposed architectures, the CNN+BiLSTM+BiGRU+Attention model achieved superior performance with an accuracy of 99.62%, precision of 100.00%, recall of 99.25%, and an F1-score of 99.62%. The results demonstrate the potential of NLP-driven systems in automating the detection of human-factor-related risks, supporting proactive and predictive aviation safety management.