The performance results obtained at the end of training and testing
procedures on the MLR, ANN, and LS-SVM models are shown in
Table 5. Considering all the results in this table, although the MAPE
value, which is a reliable performance criterion in the literature,
give successful results in the training of ANN and LS-SVM models
such as 0.91% and 0.88% respectively. The test errors of the models
are observed as a little higher and had closer values. Despite these
two models, the training and test error rates of the MLR analysis are
observed to be quite high in all of the performance criteria. The
comparison between the test results of the models and actual values
are shown in the bar chart in Fig. 4. Although these three models
that have been analyzed in this study gave quite successful
estimations between 1975 and 1987, it can be said that all three
failed in 2008. However, the error rates of ANN and LS-SVM algorithms
have close oscillations in most of the test data.