Using near misses, artificial intelligence, and machine learning to predict maritime incidents: A U.S. Coast Guard
Peter M Madsen1, Robin L Dillon2, Evan T Morris3
1Marriott School of Business, Brigham Young University, Provo, Utah, USA.
Summary
Near-miss incident data can predict future accidents. Artificial intelligence (AI) and machine learning (ML) models accurately forecast high-risk waterways, enhancing safety for the U.S. Coast Guard (USCG) and other industries.
Area of Science:
- Maritime Safety
- Data Science
- Predictive Analytics
Background:
- Organizations increasingly use incident data to improve safety.
- Near-miss incidents are recognized as predictors of future negative events.
- Advancements in AI/ML tools enhance data analytics accessibility.
Purpose of the Study:
- To determine if near-miss data in the U.S. Coast Guard's (USCG) Marine Information for Safety and Law Enforcement (MISLE) database can predict future accidents.
- To develop AI/ML models for identifying waterways at significant accident risk.
Main Methods:
- Analysis of near-miss counts from the MISLE database to predict serious casualties.
- Development of random forest decision tree AI/ML models for waterway risk prediction.
- Implementation of an R-based monthly predictive model using historical data (2007-2022).
Main Results:
- Recent near-miss counts effectively predict future serious casualties at the waterway level.
- AI/ML models achieved prediction accuracy ranging from 92% to 99.9%.
- Models were trained on 2007-2022 data and tested on 2022 data.
Conclusions:
- Near-miss data is a valuable predictor of future maritime accidents.
- AI/ML models provide accurate risk assessments for waterway safety.
- The predictive models can support USCG prevention efforts and are generalizable to other industries.
