Popular classification methods, such as Logistic Regression or Random Forest, and more innovative approaches based on quadratic optimization show significant differences in their requirements for input data, the nature of their output and the extent of accessibility of their results. Within the framework of the research project, these differences will be systematically worked out using real and simulated data sets and quantified on the basis of suitable indicators. In particular, the question will be addressed to what extent different methods can be meaningfully combined with each other.