Improving Pure Component Property Estimation in Specialty Chemistry Using Local Estimators for Group Contribution Models

Specialty chemical products are essential for many aspects of modern life, but their development and production can be costly. To improve the efficiency of the downstream production process through computerized simulation, accurate physical property data are necessary. Our proposed method involves using a local estimator to increase the accuracy of conventional group contribution models for specialty chemical products, byproducts, and intermediates. By leveraging available data on similar molecules, the local estimator reduces the median relative absolute error in melting and boiling point temperature and vapor pressure estimation by up to 50%. This will enable faster and more accurate development of the downstream production process. The approach was verified by using different sets of published and industry data and can easily be extended to further properties.