<p>Electronic aluminum foil with extremely thin specifications is widely used in the electronic engineering field, and foil rolling flatness is a key indicator for determining the product quality. The traditional flatness control model is based on rolling theory, but its precision is limited; the flatness closed-loop control system has strong hysteresis and lacks effective flatness control. To break these limitations, this study proposes a foil rolling flatness prediction and optimization method based on machine learning. The method adopts the Local Outlier Factor (LOF) to identify outliers and extracts the key features in original dataset as the model inputs by comparing feature importance. After that, an adaptive position update strategy is introduced to balance the global and local search capabilities of Particle Swarm Optimization (PSO) algorithm. Then, the modified PSO (MPSO) algorithm is used to optimize LightGBM, and the flatness prediction model is built. Compared with Back Propagation Neural Network (BPNN) and other models, the proposed model has the highest accuracy and model generalization ability, whose <i>R</i>² reaches up to 0.972. Finally, a flatness feed-forward control strategy based on model prediction (MP-FFC) is developed. The experiment successfully realized the improvement of foil rolling flatness defects. Thus, the method proposed in this study can achieve high-precision flatness prediction and optimization control, which is of great significance for improving the flatness quality of electronic aluminum foil.</p>