This research aims to predict the complexation performance of phospholipid complex by machine learning methods. 341 drugs/phospholipid complex formulations were collected from the literature. The datasets were trained by the lightGBM method. To validate the prediction modeling, berberine (BBR) was used as the model drug to form the complex with phospholipid. Molecular dynamics simulation was used to investigate the molecular mechanism for self-aggregation of BBR in solution and BBR-phospholipid complex complexation, which well explained the experimental results.