In recent years, the development of smart phones, Internet of things and cloud computing provides massive data for the study of education big data. However, the current research on education big data is not enough, as the data is small, the processing method is simple and the scope of application is narrow. In this paper, big data technology is adopted and the massive data of multiple data sources are utilized, the model features are extracted by well-designed data processing methods, and a more general high-risk student prediction model is constructed. The research of this paper can broaden the research methods and application scope of education big data, so it has strong theoretical value and important practical value.