PREDICTION OF MULTIDIMENSIONAL TIME SERIES BASED ON GS-RSR-SVR AND ITS APPLICATION IN AGRICULTURAL ECONOMY

L. F. Wang, H. Y. Zhang, Zh. M. Yuan, H. Y. Wang, Y. G. Xie

Abstract: This paper proposes a method that creatively applies a Geo-statistics tool (GS) to complete fast and adequate order determination and introduces a novel algorithm, named Reasonable Sample Rejection (RSR) to realize rational sample selection. Then, combined with Support Vector Machine Regression (SVR), a high precision non-linear prediction method named GSRSR-SVR is proposed for multidimensional time series. The main steps of the novel method includes: 1) determine the order for the dependent variable of the training samples based on one-dimensional GS aftereffect duration (range), 2) screen the independent variables according to Leave-One-Out Cross Validation (LOOCV) based on the minimum Mean Squared Error (MSE), 3) reject some oldest training samples based on the minimum correlation coefficient of fitting absolute relative error of training sets of different rejected sizes and sample number. Three real-world datasets was used to test the effectiveness of GSRSR- SVR. The results show that GS-RSR-SVR has higher prediction precision and more stable prediction ability than MLR, ARIMA, CAR, BPNN, SVR and SVR-CAR.

Keywords: geo-statistics tool; multidimensional time series; prediction; reasonable sample rejection; support vector machine regression

Date published: 2019-01-17