Random forests for feature selection in QSPR Models - an application for predicting standard enthalpy of formation of hydrocarbons

Table 1 Summary of the results (10-fold cross validation) obtained for all the models

Feature selection	Feature selection technique	Number of variables/PC	Machine learning model	RMSE^¤	q²_cv^§
No		1485	RF^‡	50.28	0.9303
		1485	SVM^†	44.47	0.9520
Yes	PCA^*	28 PC^#	SVM^†	34.87	0.9671
	GA°	58	SVM^†	47.1	0.9391
	RF - VI^ǂ	89	SVM^†	34.1	0.9686

^¤Root Mean Square Error; ^§cross validated squared correlation coefficient; ^‡Random Forests; ^†Support Vector Machines; ^*Principal Components Analysis; ^# Principal Components; ° Genetic Algorithms; ^ǂVariable Importance using Random Forests.

ISSN: 1758-2946