<p>Most existing neural network-based control results are limited to semi-global stability. This paper investigates adaptive global predefined-time (PT) exact tracking control for uncertain high-order nonlinear systems subject to time-varying input delay (TVID) and input saturation. To overcome the semi-global limitation, a switching radial basis function neural network (RBFNN) mechanism is introduced, which employs RBFNN approximation within its effective activation region and switches to a robust compensation strategy based on state-dependent inequalities when the system state leaves this region. In addition, an input-dependent compensation signal is incorporated to provide a unified treatment of TVID and input saturation. On this basis, a novel tracking controller is developed by combining backstepping design with bounded estimation techniques. The proposed controller guarantees that the tracking error converges exactly to the origin within a user specified PT, while ensuring global boundedness of all closed loop signals, despite the presence of multiple uncertainties including unknown control coefficients, unknown nonlinear dynamics, and external disturbances. Numerical simulations are provided to demonstrate the effectiveness and robustness of the proposed approach.</p>