A preliminary study of airport freight traffic forecasting based on least squares support vector machine

Based on least squares support vector machines (LS-SVM) technique, a new method for airport freight traffic volume forecasting is proposed. The basic theory of regression LS-SVM is given, together with the brief review of SVM. The historical statistical data of airport freight traffic and pertinent indices from 1983 to 2003, which are provided by National Bureau of Statistics of China and Civil Aviation Administration of China, are used as the total data set of LS-SVM. RBF kernel function, polynomial kernel function, and linear kernel function are used as the kernel function of LS-SVM to perform freight traffic volume forecasting, respectively. After five-fold cross-validation, the LS-SVM of RBF kernel function has the best forecasting performance. Eventually, RBF function is selected as the kernel function of LS-SVM and its optimal parameters are σ ²=20 and γ=13. The maximum absolute forecasting error is 12.4 and the maximum relative forecasting error is 5.66%. The results show that LS-SVM is a promising tool to perform airport freight traffic volume forecasting.