Regression models play an important role in data analysis, enabling prediction and providing data analytic tools for a better understanding of the relationship between independent and dependent variables. Linear models often fail in real life scenarios, most systems showing a nonlinear behavior. Regression trees overcome this limitation by recursively partitioning the data into several homogeneous regions and fiting a model (constant, linear, or higher polynomial) for every region. Regression trees are generalized decision trees, predicting a continuous value instead of class labels. By their means large datasets can be modeled and easy interpreted.

Regression models find application in the process of system identification, where experimental measurements of dynamical systems are recorded and mathematical models are fitted. These models are built in model predictive controllers, which have found successful applications in may industries ranging from chemicals to aerospace industry.
In this seminar we will analyze different construction algorithms for regression trees, their split strategies and flexibility w.r.t. representing the data.