Assessing spill fire characteristics through machine learning analysis

•Fuel spill fires are a significant risk in nuclear power plants and require accurate models to quantify their consequences.•The current spill fire models have large uncertainties and do not account for all variables in the dynamic process.•This study collects past spill fire experimental data to analyze the spreading and burning behaviors and identifies the dominant parameters using machine learning models to predict the peak heat release rate.•The results quantify the importance of the fuel leakage rate or quantity, the slope of the spill surface, ignition delay time, substrate material, and fuel properties, which can be used to support fire probabilistic risk assessment (PRA).