Due to the special characteristics of competitive sports, high-level athletes are more prone to psychological problems as their mental and physical bodies are subjected to long-term psychological pressure and high-intensity stimulation challenges that are difficult for ordinary people to experience. The emergence of these problems will directly affect the athletes' competitive level and potential, and may even end their athletic careers prematurely, causing irreparable damage to themselves. This paper proposes a data-driven machine learning approach for injury prevention and treatment decision support. The advantages of the machine learning method mainly lie in the following: firstly, the whole process of machine learning, from feature selection to analytical modelling and up to prediction, is tightly focused on the data, and will not be interfered by the a priori knowledge, so it can effectively extract the unattended influencing factors which are difficult to be found in the traditional method, so that the feature selection process can be more accurately completed, and it can effectively support the scientific athlete's health management.