Cabin Safety Incident Reports for Risk Assessment, 2004-2024

8,280 global cabin abnormal event reports from 2004 to 2024 were used to train a hybrid CNN-LSTM-Attention model for intelligent risk classification. The model, proposed by Lianbin Zhou, achieved 95.01% accuracy and a 94.17% F1 score on its test set. This framework establishes a mapping from text features to risk mechanisms and hazard levels for interpretable risk quantification in civil aviation.

Use Cases

• Training text classification models for aviation safety incidents based on the described reports.
• Benchmarking machine learning models for risk assessment based on the reported performance metrics.
• Developing interpretable risk quantification systems based on the described 'text feature–risk mechanism-hazard level' mapping.

Strengths

• Dataset contains 8,280 incident reports spanning 20 years (2004-2024).
• Associated model achieved a reported test accuracy of 95.01% and F1 score of 94.17%.
• Data is shared under a permissive CC-BY-4.0 license.

Limitations

• Row count for the underlying data is unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 5.5 KB, indicating limited scope.

Provenance

Source

Lianbin Zhou via figshare.

Collection Method

Likely compiled from global cabin abnormal event reports.

Time Range

2004 to 2024

Freshness

Last updated 2026-05-27 17:39:57; freshness should be verified.

Geography

Global