ABSTRACT Digital light processing (DLP) is renowned for its precision, but the challenge lies in the identification of optimal print parameters to minimise print errors and enhance overall print accuracy. This study introduces a groundbreaking approach by integrating 'Random Forest' (RF) models with DLP printing to construct a predictive model for printing errors to achieve unparalleled precision. We conducted experiments using common commercial resins to print, resulting in a comprehensive dataset of 690 experimental datasets. Consequently, we validated this approach by fabricating Y-type microfluidic structures with a minimum feature size printing of 2 µm and herringbone mixer structures with feature sizes of 21.2 µm. We successfully controlled the average print error below 2.3 µm. The 'RF' model was also extended from the printing of microfluidics (2.5D) to complex spatial three-dimensional (3D) Triple-Periodic Minimal Surface (TPMS) structures. We have successfully achieved the printing of TPMS-Gyroid and TPMS-Diamond structures with a minimum feature size of 20 µm, while the printing errors were all kept within 1.5 µm. This study underscores the generalizability and immense potential of integrating machine learning techniques with high-precision DLP printing, offering valuable insights for future research in the realm of high-precision and expeditious 3D printing.