Based on the Gradient Boosting Decision Tree (GBDT) machine learning algorithm, this study develops a model for predicting the fusion reaction cross-section (CS) of ^99-103\rmMo^*, aiming to explore the optimal synthesis pathway for the medical isotope ^99\rmMo. The model inputs include characteristic quantities such as reaction energy, proton number, mass number, and binding energy, as well as relevant parameters calculated based on phenomenological theoretical models, with the output being the fusion reaction cross-section. It is found that the mean absolute error (MAE) between the machine learning-predicted CS and experimental values on the test set is 0.0615, which is superior to the 0.1103 predicted by the EBD2 model. On this basis, combined with the GEMINI++ program, the survival probabilities of the neutron decay channels for ^99-103\rmMo^* were calculated to derive the evaporation residue cross-section of ^99\rmMo. It is found that the evaporation residue cross-section of the 2n de-excitation channel for ^4\rmHe + ^97\rmZr at a center-of-mass energy of 18.51 MeV is 1199.80 mb, making it the optimal pathway for synthesizing ^99\rmMo. This research validates the reliability of physics-informed machine learning methods in predicting fusion reaction cross-sections and provides a reference for optimizing reaction system selection and producing medical isotopes through fusion reactions in heavy-ion accelerators.