Selective and Heterogeneous SVM Ensemble for Demand Forecasting

An accurate demand forecasting model has both academic and practical significance to supply chain management for China´s retail industry. In this paper, we proposed a novel demand forecasting model named SHEnSVM (Selective and Heterogeneous Ensemble of Support Vector Machines), in which the individual SVMs are trained by different samples generated by bootstrap algorithm and different parameters generated by grid search method in order to improve the diversity among them, and then Genetic Algorithm is employed for retrieving the best individual combination schema. Finally, SHEnSVM is applied to demand forecasting of one beer retail company. The experiment results prove the model has stronger generalization ability.