Feature Selection Data Hiperspektral untuk Prediksi Produktivitas Padi dengan Algoritme Genetika dan Support Vector Regression

Abstract

that of multispectral. However, the high-dimensional feature of hyperspectral data often cause the risk of “over fitting” when analyzed. Therefore, it is necessary to reduce the dimension of hyperspectral data, which can be done by feature selection. In this study, Genetic Algorithm (GA) and Support Vector Regression (SVR) are used to select the best band as the best feature of hyperspectral data for predicting productivity of paddy. GA selects the best band, parameter C and parameter γ for using in SVR. Parameter C is used as error penalty in SVR and parameter γ is used for kernel function in SVR. SVR uses the best band and the best parameter, which both of them are selected by GA to predict yield of paddy, then give fitness value for GA. Hyperspectral data that used in this research has been taken on June 2008 in Indramayu and Subang, while the heights of the spectral acquisition is 2000 m which is called hymap data. GA-SVR method was implemented using IDL and LIBSVM in C languange. The result shows that the GA-SVR method could select the best band and decrease the error of the prediction. The best result was shown by Hymap data using RBF kernel. The number of band usage for prediction was reduced from 109 to 48, the RMSE was reduced from 0.301 to 0.017 and the R2 = 0.99.