To address the difficulty in the real-time and accurate prediction of surface roughness during the laser drilling of titanium alloy micro-holes，a prediction method based on an improved particle swarm optimization algorithm optimized backpropagation(IPSO-BP) neural network is proposed. The IPSO-BP prediction model was established considering three key process parameters—laser power，pulse frequency，and focal position deviation—as input features and the surface roughness of micro-holes as the output. The model was trained and validated using 27 sets of simulation data，with 22 sets for training and 5 sets for independent testing. The simulation test results indicated that the maximum relative error of the prediction model for the test set was 6. 89%. Nine sets of laser drilling verification experiments were conducted to further validate the effectiveness of the model. The maximum relative error between the measured roughness and the value predicted by the model was 8. 06%. The results indicate that the constructed IPSO-BP neural network model can predict the surface roughness of laser-drilled holes with high accuracy. This method provides effective guidance for process parameter optimization， reduces experimental trial and error， and lowers processing costs.