Acessibilidade / Reportar erro

Fighting Collinearity in QSPR Equations for Solution Kinetics with the Monte Carlo Method and Total Weighting

A Monte Carlo method is used in addition to functional and individual weighting to overcome multicollinearity problems in multiple linear regression equations applied as quantitative-structure-property-relationships, allowing the estimation of correct coefficient confidence intervals. The method was applied to rate constants for the Menschutkin reaction between Et3N and EtI in mono- and di-alcohols, at 25.00 ºC. Results show that the use of our methodology produces a significant improvement upon confidence interval estimates regardless of the level of collinearity present. Addition of weighting shows additional advantages, increasing the overall consistency of the regression process.

Keywords:
Monte Carlo; multicollinearity; correlation; MLR


Sociedade Brasileira de Química Instituto de Química - UNICAMP, Caixa Postal 6154, 13083-970 Campinas SP - Brazil, Tel./FAX.: +55 19 3521-3151 - São Paulo - SP - Brazil
E-mail: office@jbcs.sbq.org.br