Temporal Validation of a Human-Error Likelihood Model for Dynamic Positioning Incidents Using IMCA-Reported Data

Abstract

Dynamic positioning (DP) operations are safety-critical maritime activities in which technical system performance, environmental conditions and human decision-making interact continuously. Previous research presented a prediction model based on binary logistic regression that estimates the likelihood of human-error-related DP incidents, identifying the percentage of thrusters online as the significant predictor, using IMCA-reported data from 2007–2015. This study evaluates whether the previously developed prediction model remains valid when applied to a more recent IMCA dataset, with DP incidents from 2016 to 2025. A retrospective documentary analysis was conducted, and 77 incidents met the inclusion criteria for statistical analysis. Human-related factors were identified as a main or secondary cause in 35 cases, representing 45.5% of the analysed sample. The original thruster-based relationship was not reproduced in the updated drilling subset, where the percentage of thrusters online did not distinguish between human-related and non-human-related incidents. Applying the previous model resulted in poor discriminatory performance (AUC = 0.541), indicating that the original relationship between thruster availability and human-related incidents was no longer observed. Additional regression analyses found no statistically significant association between the selected operational, technical or environmental variables and human-related causation. These findings suggest that historical prediction models should be re-evaluated before being applied to contemporary DP operations. They also indicate that the variables available in IMCA reports alone are insufficient to explain human-related incident causation, highlighting the need for future studies to investigate additional human and organisational factors.