Abstract: It is complicated and difficult to determine the driving factors which affect changes of cultivated land area. In order to select the effective driving factors and promote the prediction accuracy of cultivated land area, this paper proposed a method to determine the driving factors of cultivated land area relying on the minimum mean squared error (MSE) of prediction value and observed value among all combinations of independent variables based on GS-SVR. Then, taking Hunan province as a case, this paper used several time-series forecasting methods, such as SVR-CAR, LSSVM, BPNN, ARIMA and MLRR to evaluate the performance of the selected driving factors. The results showed that the optimal driving factors for cultivated land changes was the combination of urbanization level and production index of the real estate industry, and with the driving factors selected by GS-SVR, all reference methods greatly improved the prediction accuracy of cultivated land area. The proposed method has an extensive application prospect for predictions involving multidimensional time series data, such as changes of cultivated land area.