Estimation of forecast quality of the two-phase scheme for solving the problem "structure-property" / E. I. Prokhorov. //Vestnik Moskovskogo Universiteta. Seriya 1. Matematika. Mekhanika. 2013. № 5. P. 34-37 [Moscow Univ. Math. Bulletin. Vol. 68, N 5, 2013.].
The paper is focused on methods of searching for quantitative structure property relationships for predicting the activity of chemical compounds. A two-phase scheme is proposed for solving the "structure-property" problem. An estimation of the quality of prediction with the use of a two-phase scheme is obtained. In particular, it is proved that the prediction quality is improved when using non-trivial rejection rules. Results of practical tests of the proposed method for solving the "structure-property" problem are presented, which confirm the efficiency of this approach.
Key words: prediction of the properties of chemical compounds, machine learning, pattern recognition, classification, "structure-property" problem.