Data-driven investigation of ship fuel consumption integrating causal inference and hierarchical analysis

This study develops a data-driven analytical framework that integrates causal inference with hierarchical analysis to reveal the underlying mechanisms influencing ship fuel consumption (SFC). The framework combines multi-source data fusion, direct linear non-gaussian acyclic model-based causal discovery, and double machine learning with causal forests model to estimate Average Treatment Effects (ATEs), followed by interpretive structural modelling for hierarchical decomposition. Experimental results under the expanded DAG-derived adjustment specification indicate that Daily sailing hours has the strongest positive conditional effect on SFC (ATE = 2.058), followed by Main engine RPM with a positive but more uncertain effect estimate (ATE = 0.268). The hierarchical analysis further organises the directional-dependence network into interpretable structural levels, thereby illustrating possible multi-level transmission patterns among operational and environmental factors. This framework provides an exploratory and graph-informed analytical basis for interpreting directional dependencies and observed-covariate conditional effect patterns in SFC, offering cautious decision support for data-driven energy management. The source code is publicly available at: https://github.com/AdvMarTech/ship_fuel_consum_causalinfer .