Regional geomagnetic field models can provide detailed insights into the geomagnetic field and are of substantial value for precise maritime navigation and the detection of marine targets. In light of the polynomial model’s numerous advantages, including straightforward computation, rapid computation speed, and high model resolution, this study delves into the polynomial modeling method. To address the challenges of enhancing solution precision, selecting the optimal polynomial function, and determining the optimal truncation degree, it introduces an improved method based on singular value decomposition to enhance the model’s accuracy at higher degrees. A comparison of the Taylor polynomial model, the Laguerre polynomial model, and the Chebyshev polynomial model revealed the superiority of the Legendre polynomial model in terms of truncation degree, modeling accuracy, and boundary effect. In addition, by employing K-fold cross-validation, the complete dataset is effectively leveraged for both fitting and validation, thereby facilitating the identification of the optimal truncation degree for each polynomial model. The superior accuracy of the regional geomagnetic field models is validated through a comparative analysis of the calculation results derived from the EMM2017 model.