Two types of high-fitness machine learning algorithms including random forest and extreme gradient enhancement Tree were used to establish a prediction model of the content of hydrogen in hydrothermal bio-oil with high-precision and wide-range（R²>0.93） based on 243 sets of experimental data collected from literature on the hydrothermal preparation of hydrogen-rich bio-oil from wastes with high-moisture and high-lipid including dead livestock and algae to construct a model for predicting the content of hydrogen in hydrothermal bio-oil with high accuracy of prediction and strong ability of generalization， and to study in depth the laws and mechanisms of hydrothermal conversion of biomass. SHAP interpretable technology and local dependence analysis method were used to analyze the contribution， local dependence response behavior and interactive coupling law of conditions for producing hydrothermal bio-oil based on the mathematical prediction model. The results showed that the content of lipid and hydrogen in the wastes with high-moisture and high-lipid were the determining factors for the preparation of hydrogen-rich bio-oil， ranked the top two in terms of its contribution to the enrichment of hydrogen in the oil phase and significantly affected the accumulation of hydrogen in the oil phase. The content of hydrogen in the oil phase was increased with the increase of the content of hydrogen in the raw material， indicating that hydrogen-rich raw materials provide convenient conditions for the preparation of hydrogen-rich bio-oil， with an improvement effect of up to 4%. The high calorific value of raw materials was the main inhibitory factor for the hydrogen-enriched behavior of bio-oil， with an inhibitory effect of up to 4%. The coupling effect between the information of element， industry， and components of biomass was strong from the interactions among the characterized categories， but the effect of local coupling between the characteristics of raw material and operating conditions was relatively small.