Figure (5). Results from SVM using RBF kernel with all features selected, the black dots is the real value of site EUI, and the red line is predicted value.
5.3.2 Support Vector Machine Feature Selection
The 6 features selected by the RFE algorithm are: LotType, ProxCode, Oil, Water, Gas, Electricity, which are basically location, physical characteristic and dominant energy type features. With 6 selected features, the optimized parameters for the residential dataset are: epsilon: 0.0, C: 501, gamma: 0.01, from this model we have result of test score R2 as 0.11 and mean squared error as 737.52. For commercial buildings, the optimized parameters for the commercial dataset are: epsilon: 0.0, C: 2001, gamma: 0.01, and the R2 from this model is 0.08 along with 1216.09 of mean squared error. Results shown as below in Figure (6).