Experience shows that ground vibration induced by high-speed trains can reach levels that cause environmental problems, such as human annoyance, possible damage to old and historical buildings, and interruption of sensitive instrumentation and processes. In engineering, understanding the characteristics of the vibration source, path, and influence distance on various influence factors is essential. Thus, a reliable vibration measurement scheme for highly sensitive vibration areas is proposed to acquire the characteristics of ground vibration induced by passing trains. From a wide variety of field measurements and monitoring using vibration sensors, the main characteristics that affect overall vibration levels include train speed, ground condition, measurement distance, and supported structure. Based on the database of measurement results, the most widely used kernel learning algorithm, support vector machine (SVM) algorithm, is adopted in this research to predict the vibration levels induced by high-speed trains on embankments. Analysis results show that the developed SVM model can predict vibration levels with an accuracy rate of 70%-85% for four types of vibration levels, namely, overall, low, middle, and high frequency ranges. The methodology for developing an automatic prediction system for ground vibration levels induced by highspeed trains is also described in this paper.