PLS2-R Modelling to Predict Critical Temperature and Pressure of Cyclic Hydrocarbons using Geometric Molecular Descriptors

Abstract

Inferential statistics allows us to reduce the dimensionality of a heterogeneous universe with high variability and high multicollinearity without losing relevant information; in this case, 129 explanatory variables have been reduced to only two, managing to construct linear regression contribution models of the PLS2-R type for highly correlated response variables, such as Tc and Pc, with high predictive correlation 0.96 and 0.91 respectively to solve simultaneously for cyclic hydrocarbons. These models allow us to approximate thermodynamics properties by numerical resolution with thermodynamics equations and infer critical points of new cyclic hydrocarbon molecular structures by molecular reconstruction methods and unknown compounds or mixer under a method safety, reliably, and cheaply in industrial applications, environmental studies, and health risk assessments.