Improving model predictions for RNA interference activities that use support vector machine regression by combining and filtering features

Table 1 Feature mapping methods performance in RBF-epsilon regression SVM model training and testing within dataset₂₄₃₁

				train₂₄₃₁
				test₂₄₃₁ 10 × cross validation
Feature mapping method	FN₂₄₃₁	R	MSE	R	MSE
1-Position specific base	84	0.784	0.016	0.711	0.026
2-Thermodynamics	23	0.915	0.007	0.640	0.029
3-Entropy	23	0.730	0.021	0.094	0.046
4-Guide strand structure	24	0.430	0.033	0.293	0.041
5-Guide strand features	32	0.266	0.037	0.243	0.042
6-N-Grams N = 2	16	0.408	0.033	0.291	0.041
7-N-Grams N = 3	64	0.656	0.024	0.435	0.037
8-N-Grams N = 4	256	0.590	0.027	0.532	0.034
9-N-Grams N = 5	1024*	0.590	0.029	0.487	0.036
10-N-Grams N = 6	4096*	0.621	0.036	0.439	0.036
11-N-Grams N = 2–5	1360*	0.614	0.026	0.559	0.033
12-Target strand structure-nondirectional	22	0.646	0.024	0.257	0.045
13-Target strand structure-directional	43	0.607	0.025	0.277	0.042
14-Target imprecise thermo	22	0.932	0.007	0.272	0.045

FN_dataset = Feature Number count from dataset, R = Pearson correlation coefficient, MSE = mean squared error * Theoretical, but five 5-grams and several 6-grams are absent in the present dataset, reducing the effective feature set size.