Vibrations induced by the operation of underground trains result in certain changes in the flow characteristics of the underground soft soil, which may lead to problems like ground settlement and damages of subway tunnels. In this study, an improved drag-sphere device is implemented to investigate the flow ability of soft soil subjected to vibrations, and the experimental results indicate that vibrations with high frequencies and low confining pressure enhanced the flow ability of soil samples. Then an artificial neural network (ANN) model is developed based on the obtained experimental data to predict the soil viscosity, where the genetic algorithm (GA) is implemented to optimize the weights and biases in the network. Specifically, by comparing the simulated results with experimental data, the optimal topology, training algorithm, and transfer functions are selected for the proposed model, and the model predictions are in high agreement with the experimental data, which denotes the proposed ANN model is accurate and reliable. Moreover, an analysis on the contributions of each input reveals that the water content affects the soil viscosity most while the frequency has the least impact for a single factor, which is in correspondence with the fact that the flow ability of soft soil is mainly affected by the geological conditions and its natural properties.