Sparse kernel machines based on type-2 fuzzy clustering for demand forecasting

Abstract

In this paper, a new method for demand forecasting of motor vehicle units is proposed. The method is based on interval type-2 possibilistic fuzzy C-means clustering and a special type of sparse kernel machines known as support vector regression. Interval type-2 possibilistic fuzzy C-means clustering is used to partition the input space. Then, indicator variables are selected. For each cluster of input data sets, support vector regression is applied to estimate future demands of motor vehicles. Least square method is used for variable selection of the model. Three other methods including pure support vector regression (Pure-SVR), least square support vector regression (LS-SVR), and interval type-2 fuzzy C-means support vector regression (IT2FCM-SVR) are used to validate the proposed model. Finally, the RMSEs of all sub-problems are compared to each other. The results of calculating RMSEs for all methods show that the proposed method (IT2PFCM-SVR) outperforms other techniques used in this paper, and it significantly reduces forecasting error.