A TOPSIS-Based Regression Fusion Model for Prediction in Maritime Evacuation

In emergency evacuation scenarios, obtaining accurate evacuation time is crucial for developing effective evacuation plans and supporting emergency response. This study aims to construct a framework for evacuation time prediction in maritime emergencies. It uses a Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) based Regression Fusion (TRF) model to fuse predictions from a set of advanced Machine Learning (ML) models, and evaluates its performance using multi-criteria decision making with an emphasis on prediction accuracy and overall comprehensiveness. First, an agent-based modelling technique is used to simulate the evacuation process, and a multi-scenario dataset is generated from the simulation outputs. Second, seven ML models are trained for predicting evacuation time, and four complementary evaluation metrics are used to compare model performance. Third, TOPSIS is used to calculate integrated scores from the multi metric results, and these scores are converted into fusion weights for TRF. Finally, TRF fuses the predictive time from the seven ML models into a single integrated output, which can retain complementary strengths across models and improve overall predictive performance. The results reveal that TRF achieves lower prediction errors and more stable performance than the individual models across scenarios. This study provides an insightful scientific basis for using integrated ML prediction tools to support evacuation planning and safety assessment for maritime evacuation.

Keywords: Maritime safety, Emergency evacuation, Passenger ship, Evacuation simulation, Machine learning, TOPSIS.