Municipal revenue prediction by ensembles of neural networks and support vector machines

Municipalities have to to pay increasing attention to the importance of revenue prediction due to fiscal stress. Currently, judgmental, extrapolative, and deterministic models are used for municipal revenue prediction. In this paper we present the designs of neural network and support vector machine ensembles for a real-world regression problem, i.e. prediction of municipal revenue. Base learners, as well as linear regression models are used as benchmark methods. We prove that there is no single ensemble method suitable for this regression problem. However, the ensembles of support vector machines and neural networks outperformed the base learners and linear regression models significantly.