Recently, increasing system complexity and various customer demands result in the need for highly efficient vehicle development processes. Once the brake torque is predicted accurately during the driving scenario in the earlier stage, it will be able to prevent the changing the vehicle or brake system design to satisfy the legal regulation and customer requirement. As brake torque performance target allocate brake pad friction coefficient level and characteristic, the accurate friction coefficient prediction should be preceded for accurate prediction for brake torque. Generally, the friction coefficient of the brake pad is known to vary nonlinearly depending on the physical properties of the disc and the pad, as well as the brake disc rotational speed, the disc temperature, and the hydraulic pressure. Furthermore, it varies depending on the driving scenario even when other conditions are the same. Therefore, it is necessary to apply new methods to solve these challenges. In this study, new derivative variables are discovered by exploring the feature that is highly related with the brake pad friction coefficient. And then, MERF algorithm is selected to train the machine-learned meta model for consistently excellent predictive performance in the various driving scenarios. As a result, the overall predictive performance has been improved and achieved the target MAE. In addition, it was confirmed that the offset, discontinuity, and noise vulnerability of the prediction results also improved compared to previous studies. Most of all, it has been explainable that the causes of the brake pad friction coefficient changing. This meta model can predict the dynamic brake pad friction coefficient with changing brake disc rotation speed, disc temperature and hydraulic pressure input. It is also possible to predict the braking performance with varying friction coefficient according to the driving scenario when the meta model is integrated to the total vehicle simulation model.