Evaluating the welding joint quality in real time is difficult for chassis parts robotic gas-shielded welding. Series of metal active gas (MAG) joints were conducted in this paper to investigate the relationship between welding current, welding speed, energy input, and weld bead geometry. Bead width and bead reinforcement are obtained using a line-structured light measurement method, and the penetration depth of the bead is measured with the macroscopic metallurgical microscope. The ratio of penetration depth to the plate thickness and reinforcement is chosen as the evaluation criterion of the joint quality. Based on XGBoost algorithm, two data-driven models are proposed to recognize penetration status and predict the bead reinforcement. In the prediction results, the absolute error of the penetration coefficient is 0.079 at the maximum, and the average relative error is 11.06%. For the test result of reinforcement prediction model, the relative error is 20.5% on average. The test results show that the XGBoost-based models can be used for real-time prediction of welding quality.