Predicting dropout rates accurately on digital education platforms such as ProFuturo will enable preventive measures to be taken, significantly reducing dropout rates and optimizing the use of resources. Several authors have addressed this problem using machine learning and artificial intelligence techniques with encouraging results. However, most of the approaches are based only on demographic variables or the course completion certificate discarding relevant information available in Moodle platform.Furthermore, they obtain moderate success rates and are not easily interpretable in terms of the indicators considered.In this paper we propose a novel methodology for accurate dropout prediction that takes into account all informative variables fromMoodle. The approach is based on simple machine learning models and maintains high interpretability in terms of input variables.The experimental results show that a methodology based on Random Forest can achieve high detection probability (91%) without compromising specificity with a value of 88%. Moreover, the application of SHAP algorithm has provided high interpretability to understand the role of different variables.