Internet of things-driven approach integrated with explainable machine learning models for ship fuel consumption prediction

Abstract

The International Maritime Organization has proposed several operational policies and measures to lower ships' specific fuel consumption (SFC) and associated emissions toward the sustainability of maritime activities, showing the need for creating exact predictive models based on actual operational conditions. Modern combined and integrated techniques between highly precise sensors, the Internet of Things (IoT), and advanced machine learning (ML) can help in accurate real-time data collection and robust prediction model building. In this work, an IoT-driven approach combined with explainable ML models was developed to predict the SFC of ships based on data collected from high-quality sensors. Indeed, five different MLs were employed including linear regression, decision tree, random forest, XGBoost, and Gradient Boosting Regression. Resultantly, XGBoost emerged as the best model for predicting SFC with the highest R² (Train: 0.997, Test: 0.95), lowest MSE (Train: 1.052, Test: 16.791), and minimal MAPE (Train: 0.08 %, Test: 0.23 %). Moreover, the interpretability analysis identified "Main engine shaft power" as the most significant predictor with a mean SHAP value of around 3.5. More importantly, these findings highlighted the importance of engine power, torque, and speed in driving model predictions for ship SFC, thus helping in a comprehensive understanding of the black-box model. • The Internet of Things was employed for data collection from multiple sensors. • Five machine-learning techniques were developed to predict ship fuel consumption. • SHAP analysis and LIME were used for explainable machine learning. • XGBoost was superior in both prediction accuracy and low errors. • Main engine shaft power was found as the most significant predictor.

Topics & Concepts