Application of random forests method to predict the retention indices of some polycyclic aromatic hydrocarbons

• The QSPR modeling was developed using RF, MLR and ANN to predict the RI of some PAHs.
• The key parameters in RF including the ntree and mtry were optimized simultaneously.
• RF has no overfitting problem and does not require a variable selection in advance.
• The performance of the RF model was superior to those for the MLR and ANN ones.

In this work, a quantitative structure–retention relationship (QSRR) investigation was carried out based on the new method of random forests (RF) for prediction of the retention indices (RIs) of some polycyclic aromatic hydrocarbon (PAH) compounds. The RIs of these compounds were calculated using the theoretical descriptors generated from their molecular structures. Effects of the important parameters affecting the ability of the RF prediction power such as the number of trees (nt) and the number of randomly selected variables to split each node (m) were investigated. Optimization of these parameters showed that in the point m = 70, nt = 460, the RF method can give the best results. Also, performance of the RF model was compared with that of the artificial neural network (ANN) and multiple linear regression (MLR) techniques. The results obtained show the relative superiority of the RF method over the MLR and ANN ones.