Abstract: The wellbore trajectories of offshore extended reach wells are complex and characterized by large horizontal displacements, leading to increased downhole friction and subsequently affecting drilling efficiency. This paper introduces a novel method for rate of penetration prediction and drilling parameter optimization in extended reach wells using machine learning, based on drilling and logging data. Initially, raw field data were preprocessed and subjected to correlation analysis, revealing significant correlations between drilling parameters such as bit pressure and rotary speed, as well as wellbore trajectory parameters like hole deviation angle and horizontal displacement, with rate of penetration. Based on these findings, rate of penetration prediction models were developed using BP neural networks, random forests, and support vector machines. The prediction accuracy of these models was evaluated using four performance indicators, with the results showing that the BP neural network model outperformed the others, providing relatively accurate rate of penetration predictions for offshore extended reach wells. Furthermore, the Bayesian optimization algorithm was employed to adjust controllable parameters such as bit pressure, rotary speed, and pump rate, resulting in an average increase in rate of penetration by 18.86%. This study elucidates the impact of drilling parameters and wellbore trajectory parameters on rate of penetration; in extended reach wells and provides theoretical evidence for enhancing drilling efficiency.