Aviation Incident Reports for Classification, 53,770 Summaries Spanning 80 Years

53,770 aviation occurrence summaries from the Transportation Safety Board of Canada spanning 80 years. The dataset was used to develop and compare three machine learning classifiers for automated incident/accident classification. The work was authored by Sawera Qureshi and published on figshare in May 2026.

Use Cases

• Automated triage of aviation safety reports based on textual summaries
• Identification of potentially mislabeled incident/accident reports requiring expert review
• Support for severity assessment pipelines in high-volume aviation safety operations
• Benchmarking machine learning models for text classification on aviation safety data

Strengths

• Contains 53,770 aviation occurrence summaries, providing a substantial corpus for analysis
• Covers an 80-year time range, offering historical depth
• Used in a study achieving a classification accuracy of 98.06% with a Support Vector Machine model

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• The dataset is very small (9.5 KB), suggesting limited scope or highly compressed text

Provenance

Source

Transportation Safety Board of Canada

Collection Method

Likely compiled from official aviation occurrence reports

Time Range

Spanning 80 years (specific start and end years not provided)

Freshness

Last updated 2026-05-21 17:44:27; freshness should be verified