The flexible, changing, and uncertain nature of present-day society requires its citizens have new personal, professional, and social competences which exceed the traditional knowledge-based, academic skills imparted in higher education. This study aims to identify those factors associated with active methodologies that predict university students' learning achievements in a digital ecosystem and thus, optimize the learning-teaching process. The teaching management tool Learning Analytics in Higher Education (LAHE) has been applied to a 200-student non-probabilistic incidental sample spread over 5 different university courses, enabling a personalized learning-teaching process tailored to the needs of each group and /or student. Based on a pre-experimental design without a control group, an analysis through decision trees based on educational data mining has been undertaken on the predictive potential of the active methodologies employed, and their effects on students' learning achievements. The criterion variable of the study was the final exam grade, and the explanatory variables included student characteristics, indicators of the teaching--learning process and non-cognitive factors. Results show that factors associated with active methodologies correctly predict a significant portion of the learning achieved by students. More specifically, the factors that have the greatest impact on learning are those related to academic engagement and to a student continuous learning process.