Objective Since vehicle dynamics models are seldom employed in data-driven trajectory prediction research this study investigates and resolves a series of critical issues including how such models can improve prediction accuracy and clarifying the unclear mechanisms through which dynamics models influence trajectory prediction precision. Methods A composite vehicle model was constructed using the vehicle lateral dynamics model and a two-degree-of-freedom kinematic model to preprocess the NGSIM dataset. After obtaining data incorporating vehicle dynamics model information the unscented Kalman filter UKF model was employed to predict vehicle trajectories. Finally the impact of vehicle lateral dynamics model parameters on the trajectory prediction accuracy of the UKF model was analyzed based on experimental results. Results The composite model significantly improved accuracy across all time periods. When combined with the kinematic model during the 0~1 s interval simultaneously using β and either θ or γ increased accuracy. During the 1~2 s and 0~2 s intervals using γ decreased accuracy while simultaneously using β and θ improved accuracy. Conclusion Compared with the pure kinematic model the relevant parameters of the vehicle lateral dynamics model exhibit different impacts on the trajectory prediction accuracy of the UKF model at different time periods.Integrating the complete lateral dynamics model with the kinematic model significantly enhances trajectory prediction accuracy across all time periods and provides better stability in accuracy.