Logo image
Algorithmic Aircraft Accident Analysis Using Machine Learning Models for Predictive Risk Management
- Paper    

Algorithmic Aircraft Accident Analysis Using Machine Learning Models for Predictive Risk Management

John Olalere Ogunlola, Olayinka Anthony Ojo, Opeyemi Victor Omolade, Ogechi Smart Ekejiuba, Babatunde Alexander Abiola Bartholomew Sunday Alfred
2026 International Conference on Cybersecurity, Digital Forensics, and AI Applications (ICCSDFAI), pp.276-283
2026 International Conference on Cybersecurity, Digital Forensics, and AI Applications (ICCSDFAI) (Istanbul, Turkiye, 25/06/2026–27/06/2026)
25/06/2026
: https://doi.org/10.1109/ICCSDFAI70505.2026.11647928
accident Aircraft Artificial intelligence aviation gradient boosting Labeling Machining Modeling Training Accidents Aging Machine Learning Safety
This study presents a decision-support framework for aviation safety analytics using National Transportation Safety Board (NTSB) accident records from 2015-2025 and a temporally validated, leakage-controlled benchmark for aviation risk modelling. Temporal, geographic, technical, operational, and environmental features were used to evaluate several machine-learning models for multiclass accident severity classification. Performance was assessed using class-sensitive metrics including Macro-F1, Weighted-F1, Macro ROC-AUC, precision, and recall. Among the evaluated models, Gradient Boosting achieved the strongest class-balanced severity performance with Macro-F1 = 0.3836 and Macro ROC-AUC ≈ 0.6607. In addition, an interpretable RiskScore was developed using pre-event variables and used to construct tri-class (Low/Medium/High) and binary high-risk alerting formulations. These tasks produced substantially higher classification performance, reflecting the structured relationship between the RiskScore-derived labels and the underlying input variables. Consequently, these results should be interpreted as evidence of effective risk stratification and screening within the proposed framework rather than as independent prediction of future accident outcomes. The primary contribution of the study is a transparent and temporally validated framework that combines conventional machine-learning models with an interpretable risk scorecard for aviation safety analysis. The findings suggest that machine-learning methods can support the identification of risk-relevant patterns in historical accident data and may provide useful decision-support information for aviation safety management when used alongside human oversight, calibration, and continuous monitoring.

(2)

pdf
Algorithmic Aircraft Accident Analysis Using Machine Learning Models for Predictive Risk Management885.55 kB
Accepted Open Access  — This license enables reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator.
url
Link to Published Version
Published (Version of record) Publisher may require payment for access  — This item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation (such as any Fair Dealing allowances allowed by copyright law). For other uses you need to obtain permission from the rights-holder(s).
1
3
Logo image

Usage details for all content viewed and downloaded in this site are shared with IRUS-UK (Institutional Repository Usage Statistics UK). Cookies are used to remember your decision. Click Accept to accept usage details sharing and the cookies.