Predictive Performance Assessment in Simulation Training using Machine Learning

Abstract Maritime simulators are a central tool for the education and training of navigators, allowing them to develop and improve their skills in a controlled and replicable environment. Despite efforts to enhance the simulation training performance assessment, there are few reliable approaches to take advantage of readily available data from simulator logs to inform performance evaluation and training adjustments. Harnessing this data more effectively could enhance the way we assess simulation training and provide a more transparent understanding of learning progress and areas for improvement. To develop a learning analytics dashboard (LAD) for performance assessment in maritime simulation training, we analyse simulator log data with 27 potential input features to predict student performance as the target feature. After filtering down to 13 potential input features using data visualization and expert validation, a cloud artificial intelligence platform is used for predicting student performance. A total of 58 algorithms were trained, of which the eXtreme Gradient Boosted Trees Classifier algorithm is adopted for prediction. The results demonstrate the potential for utilizing machine learning algorithms in analysing maritime navigation training data paving the way for a new direction in simulation training assessment.