Artificial neural network study for the estimation of the C-H bond dissociation enthalpies

The three-layered feed-forward type artificial neural network (ANN) was applied to estimate the bond dissociation enthalpies (BDEs) of the C-H bonds in partially halogenated alkanes and ethers. The training values consisted of BDEs of the 93 C-H bonds in 53 alkanes and 24 ether molecules, which were computed by using density functional theory at the (RO)B3LYP/6-311G** level. As input parameters of ANN, we defined 14 kinds of topological descriptors, such as H-C...X (X = H, F, Cl, Br, O), H-C-C...X (X = H, F, Cl, O), H-C-C-C...X (X = H, F, O), H-C-O-C...X (X = H, F). Consequently, the average of absolute deviation in correlation and estimation were 0.70 and 0.84 kcal/mol, respectively. It was possible to construct a reliable and useful BDE estimation method. (C) 2002 Elsevier Science B.V. All rights reserved.