To address the issues of insufficient objectivity and difficulty in locating muscle dysfunction in current lower-limb gait assessment methods, this paper proposes a personalized lower-limb gait assessment method that integrates a musculoskeletal model with machine learning. First, a personalized musculoskeletal model is constructed using human biomechanical simulation software through model scaling and optimized calibration of key muscle parameters, and the validity of the proposed model is verified by comparing simulation results with experimental data. Second, the peak coefficient of variation (PCV) is introduced to quantify the stability of muscle control, and differences in neuromuscular control patterns of the lower limbs between healthy individuals and stroke patients are analyzed. Finally, a gait assessment model is developed using machine learning to recognize gait abnormalities and identify abnormal muscles, combined with rule-based reasoning to generate personalized rehabilitation training plans. Experimental results show that the output of the proposed gait assessment model is highly consistent with clinical grading results, achieving an accuracy of 92.3% and a Kappa coefficient of 0.87. The findings confirm that the proposed lower-limb gait assessment method can effectively quantify the gait and muscle characteristics of stroke patients, enabling both quantitative and qualitative evaluation of gait, thus providing an objective basis for personalized rehabilitation of stroke patients.