SiC MOSFETs are widely used in power electronics converters due to their superior performance. In order to evaluate the switching performance, analytical models are widely used due to their fast calculation speed. However, the high-order and nonlinear characteristics introduced by inaccurate or unconsidered parameters are hard to be modeled. Therefore, this paper proposes a compensated analytical model for SiC MOSFET to enhance the modeling accuracy. The proposed model comprises two parts: the analytical model part for basic accuracy, and the neural network part for compensation. The neural network is trained using data from a few working conditions to fit the deviations between the experimental results and the analytical model calculations. Furthermore, expert knowledge is integrated into the design and the training process to reduce the training stress. Verification results show that with the well-trained neural network, the compensated analytical model can precisely predict the switching waveforms of the unlearned working condition with only a small increase of computational time.