This study analyzes the sentiment of final cockpit conversations before aviation accidents using the BERT (Bidirectional Encoder Representation from Transformers) method. The conversation transcripts were obtained from the Cockpit Voice Recorder (CVR) and categorized into four sentiments: Neutral, Positive, Negative Stress-Panic, and Negative Frustration. The data were labeled and processed to train the BERT model and evaluated using confusion matrix and classification report. The results indicate that negative stress-panic was the most dominant sentiment in the final conversations before the accident, with high accuracy (precision 0.96, F1-score 0.98), followed by a lower frustration sentiment score. An eval-loss value of 0.28 suggests the model is stable and not overfitting. This study provides valuable insight into the emotional state of pilots in critical situations and highlights the potential for developing cockpit stress detection systems to enhance aviation safety.
Unraveling Emotions in the Cockpit: BERT-Based Sentiment Analysis of Pilot Communications Before Accidents
Advances in Engineering res
Adisutjipto Aerospace, Science and Engineering International Conference ; 2024 ; Yogyakarta, Indonesia December 11, 2024 - December 12, 2024
Proceedings of the Adisutjipto Aerospace, Science and Engineering International Conference (AASEIC 2024) ; Chapter : 4 ; 20-27
2025-03-29
8 pages
Article/Chapter (Book)
Electronic Resource
English
Cockpit updates target CFIT accidents
Online Contents | 2010
Attitudes, emotions, accidents
Engineering Index Backfile | 1959
|Cognitive cockpit systems: information requirements analysis for pilot control of cockpit automation
British Library Conference Proceedings | 2001
|