The eight studies compiled in this thesis address various topics on the phenomenon of dropping out of higher education, determinants leading to study success or dropout, and the private returns to education. A comprehensive review of the student dropout phenomenon is provided along with an overview of definitions, theoretical models, and dropouts perspectives in the first two studies. The third paper deals with the operationalisation and the empirical examination of the ideal-typical determinants affecting student dropout decisions. In the fourth study, the dropout decision is modelled using a machine learning technique, namely random forest, and the focus is put on the very early prediction of student dropout. The fifth study aims to develop from several machine learning models the optimal classifier capable of accurately predicting potential dropout student, providing a straightforward interpretation of the results and an easy implementation. The sixth paper provides a detailed analysis of the different dropout motives and identifies the distinctive types of dropout students using cluster analysis. In the last two papers, returns to various education levels are examined, and estimates on the private rates of returns to education are provided. From an institutional and individual point of view, dropouts may point to inefficient use of resources by universities, low-quality teaching, loss of reputation as well as a personal failure, and both wasted time and monetary investments. Therefore, universities need to search for measures to counteract dropouts. The measures provided in this thesis include information offerings, study counselling opportunities throughout the entire study programme and mentoring programmes.