In recent years, perovskite-based solar cells have demonstrated outstanding performance in terms of efficiency, attracting significant attention from researchers. However, the fabrication process of perovskite solar cells typically relies on traditional trial-and-error methods to optimize process parameters, which is a time-consuming and inefficient process. To address this issue, this paper proposes a machine learning-based strategy for optimizing process parameters, using random forest, extreme gradient boosting, and adaptive boosting algorithms as base learners. A perovskite solar cell efficiency prediction model (PCEPM) was constructed using a weighted averaging ensemble strategy to predict the efficiency of perovskite solar cells for different process parameters. Experimental results demonstrated that PCEPM performance was excellent in predicting perovskite solar cell efficiency with root mean-squared error of 0.620, mean absolute error of 0.469, and R² of 0.838. Furthermore, by predicting randomly generated process parameters and selecting the optimal ones for experimental validation, perovskite solar cells were successfully fabricated with an efficiency of 23.72%, which is a significant improvement in research and development. This approach effectively uncovers the relationships between process parameters, reduces optimization time, and provides a new perspective for the application of machine learning in perovskite materials development.