Municipal revenue prediction by support vector machine ensembles
Petr Hájek, Vladimír Olej · Annual Conference on Computers · 2010
Fiscal stress has forced municipalities to pay increasing attention to the importance of revenue prediction. Currently, econometric models and expert opinions are used for municipal revenue prediction. In this paper we present a design of support vector machine ensembles for the prediction of municipal revenue. Linear regression model and feed-forward neural network ensembles are used as benchmark methods. We prove that stochastic gradient boosting outperforms the other methods when creating SVM ensembles for this regression problem. Further, bagging shows best performance for feed-forward neural network ensembles, and dagging is preferable for linear regression model ensembles.