The paper introduces CoBDock, a novel machine learning-based consensus blind docking method designed to improve the accuracy and reliability of predicting binding sites and poses for small molecules on proteins. Traditional blind docking methods, which explore the entire protein surface, often suffer from low accuracy due to the large search space. Cavity detection-guided methods, which use tools like P2Rank and Fpocket to identify potential binding sites, have shown improved accuracy but are highly dependent on the quality of the cavity detection tool. To address this limitation, CoBDock integrates multiple docking algorithms (AutoDock Vina, GalaxyDock3, ZDOCK, and PLANTS) and cavity detection tools in parallel, leveraging machine learning to aggregate their results. This approach enhances the robustness and accuracy of binding site identification and pose prediction. The method is evaluated on several datasets, including PDBBind 2020, ADS, MTI, DUD-E, and CASF-2016, demonstrating superior performance compared to state-of-the-art methods like Fpocket, P2Rank, CB-Dock, and CB-Dock2. CoBDock's effectiveness is further validated through benchmarking, showing significant improvements in binding site accuracy and pose prediction accuracy.The paper introduces CoBDock, a novel machine learning-based consensus blind docking method designed to improve the accuracy and reliability of predicting binding sites and poses for small molecules on proteins. Traditional blind docking methods, which explore the entire protein surface, often suffer from low accuracy due to the large search space. Cavity detection-guided methods, which use tools like P2Rank and Fpocket to identify potential binding sites, have shown improved accuracy but are highly dependent on the quality of the cavity detection tool. To address this limitation, CoBDock integrates multiple docking algorithms (AutoDock Vina, GalaxyDock3, ZDOCK, and PLANTS) and cavity detection tools in parallel, leveraging machine learning to aggregate their results. This approach enhances the robustness and accuracy of binding site identification and pose prediction. The method is evaluated on several datasets, including PDBBind 2020, ADS, MTI, DUD-E, and CASF-2016, demonstrating superior performance compared to state-of-the-art methods like Fpocket, P2Rank, CB-Dock, and CB-Dock2. CoBDock's effectiveness is further validated through benchmarking, showing significant improvements in binding site accuracy and pose prediction accuracy.