Interval support vector regression enables high-throughput machine learning predictions for dielectric constant of polymer dielectrics

Abstract

Accurate and rapid prediction of dielectric constant (ε) for polymer-based dielectrics at various frequencies remains challenging. We construct a dataset of dielectrics with an easily attainable numerical representation scheme. We propose an interval support vector regression with a particle swarm optimization to accelerate the ε prediction, discovery, and design of polymer dielectrics at various frequencies (spanning from 100 Hz to 1015 Hz). The key features affecting dielectric constant could be identified, and this is highly valuable to target the discovering of polymer dielectrics as high-throughput screening and tailor the desirable property.