Porous biochar contains rich multi-scale pore structure, which makes it have excellent CO2 adsorption performance. To address the shortcomings of the traditional CO2 adsorption prediction models built with experimental data, such as low accuracy and complicated calculation, this paper adopts machine learning methods such as gradient boosting decision tree (GBDT), extreme gradient enhancement algorithm (XGB), and light gradient booster algorithm (LGBM) to make model predictions of CO2 adsorption by biochar, and conducts a comparative analysis of the prediction results. The results showed that the top three factors affecting CO2 adsorption were the specific surface area, C content, and O content of biochar in order. all three algorithms could effectively predict the CO2 adsorption performance of biochar. In comparison, LGBM has the highest prediction accuracy of 94%; GBDT has significant advantages for anomalous sample data processing; and XGB has more stable prediction results for different test set variations. The results of this model can provide important references for regulating and optimizing the composition and structure of biochar.