Traditional building seismic design has problems of complex parameter selection and difficult optimization. This paper introduces an optimization algorithm based on machine learning, aiming to improve the reinforcement effect by accurately predicting and automatically optimizing the design parameters of building seismic reinforcement. Firstly, the seismic performance data of multiple actual buildings are collected and sorted, and a multiple regression analysis model is used to establish a seismic effect prediction model under different reinforcement schemes. The model is trained through machine learning algorithms to identify key design parameters that affect the seismic performance of buildings and provide data support for subsequent optimization. Then, combined with PSO (Particle Swarm Optimization), multiple parameters in building reinforcement design are searched, and the fitness function is set to evaluate the seismic capacity and economy of each solution, automatically finding the optimal solution. Finally, verification experiments are carried out using a variety of building models to simulate the seismic performance of different reinforcement schemes. By comparing and analyzing the performance changes before and after optimization, the actual application effect of the method is evaluated and its feasibility in actual engineering is verified. The optimized design increased the wall thickness to 28 cm and selected reinforced concrete materials, which increased the earthquake resistance to 85 points. The optimized design not only significantly improved the earthquake resistance but also shortened the construction period to 10 months. The machine learning-based building seismic reinforcement design parameter optimization method proposed in this study can effectively solve the optimization difficulties and inefficiency problems existing in traditional design methods, and has high practicality and feasibility.