Predictive of dielectric constant in polymers by Quantitative Structure-Property Relationships (QSPR)
Description
The dielectric constant (ε) reflects the ability of a material to align and orient electrical dipoles within its structure in response to an externally applied electric field; the greater the polarizability of molecules, the greater the value of dielectric constant. The study used a data set of 86 polymers to develop a structure-property quantitative relations (QSPR) model to predict the dielectric constant in polymers. A set of 1273 descriptors was used to construct two Gradient Boosting Regressor models (GB_A and GB_B). The best performing model (GB_A) with 8 descriptors, exhibited a performance of (R2train) = 0.938 and (R2test) = 0.802. The models were internally validated by 5 folds cross-validation, demonstrating robustness. Additionally, using the Accumulative Local Effect (ALE) technique, we analyzed the relationship between the 8 descriptors involved and how dielectric constant predictions of the model impact
Files
Steps to reproduce
Authors acknowledge the support from the National Science Foundation under grant number NSF CHE-1800476. This work is also supported in part by the NSF MRI award OAC-2019077, as well as ND EPSCoR award #IIA-1355466 and by the State of North Dakota. The authors thank Prof. Paola Gramatica for generously providing a free license for the QSARINS software. Authors also thank the Extreme Science and Engineering Discovery Environment (XSEDE) for the award allocation (TG-DMR110088). Supercomputing support from CCAST HPC System at NDSU is acknowledged. Moreoverthis, work was supported in part by Grant IKERDATA 2022/IKER/000040 funded by NextGenerationEU funds of European Commission.