Strong and frequent earthquakes have increasingly highlighted the threats posed by soil liquefaction to the safety of engineering facilities, and the mechanisms of sand liquefaction as well as its prediction and evaluation methods have become central issues of ongoing concern in earthquake engineering. A shaking table test of saturated fine sand foundation under near-field and far-field horizontal earthquake motion was carried out, and the macroscopic phenomena, lateral deformation, acceleration, pore pressure ratio, etc. of the model foundation soil were analyzed. The results show that the liquefiable model foundation sinks under the action of earthquake, cracks are generated locally after the earthquake, and there is obvious sandblasting and water bubbling phenomenon; the foundation undergoes shear deformation laterally, and the peak displacement shows obvious asymmetry. The saturated fine sand layer is more sensitive to the low-and medium-frequency part of the seismic wave; during small earthquakes, the acceleration amplification effect from bottom to top in the foundation is not obvious; during medium and large earthquakes, the acceleration of the saturated fine sand layer does not show obvious amplification, while the clay layer has a more obvious amplification phenomenon; in the early stage of the earthquake, the pore pressure is temporarily negative due to the shear dilatancy effect of the soil, and then rises rapidly, and finally the pore pressure gradually dissipates. The traditional liquefaction identification method is greatly affected by human factors. Based on SVM(support vector machine), XGBoost(extreme gradient boosting) and LR(logistic regression) machine learning models, the thickness of the overlying soil layer, epicenter distance, soil acceleration and earthquake intensity are selected as input parameters for liquefaction prediction. Among them, the XGBoost model has the best effect, with a determination coefficient of 0.98. The XGBoost-based machine learning model can be used as an effective supplement for the initial judgment of sand liquefaction prediction.