Learning Piecewise ARX Models via Regression Trees with Probabilistic Guarantees

Recent research literature shows that system identification techniques can be successfully combined with machine learning to improve the accuracy of the models obtained. In this context, the contribution of this work builds upon a research line that combines the Regression Trees method with AutoRegressive eXogenous identification to derive models of dynamical systems exploiting historical data. The main contribution of this paper is to formally relate such methodology with the scenario approach framework, thus providing probabilistic guarantees on the derived model. The proposed method is validated on a real experimental setup: first a comparison in terms of accuracy with the former method - which does not provide probabilistic guarantees - is provided, then the effectiveness of the derived probabilistic guarantees is validated on the testing dataset from our experimental setup.