The problems caused by defects in complex castings are particularly serious in automotive core part manufacturing and other key areas, which makes it urgent to predict the defects of complex castings and improve their production quality. In this paper, aiming at the problem of serious imbalance in the production data of complex 6DM cylinder block castings, such as those of pores and sand holes collected during the actual casting process, the defect prediction of complex 6DM cylinder block castings based on data synthesis and machine learning was studied, and the research status of artificial neural networks and complex casting defect prediction was combed. Combined with the on-site production situation of enterprises, demand analysis was carried out, and the production data of 6DM cylinder block complex castings were obtained. The synthetic dataset created based on the synthetic minority oversampling technique (SMOTE) algorithm was adopted as the dataset of the training model, which achieved a prediction accuracy of 99.37%. The results show that the constructed defect prediction model can accurately predict the defects in complex castings.