Machine learning and dimensional analysis assisted predictive models

Abstract

The success of machine learning approaches to high throughput screening for drug discovery in the pharmaceutical industry has inspired similar schemes for materials discovery [8, 13]. One of the challenges in adopting a datadriven approach to materials discovery is the lack of large experimental data, especially in the space of inorganic materials. To overcome this problem, hybrid approaches have been developed in which properties calculated from crystal structure by ab initio density functional theory are combined with experimental data [18]. These hybrid approaches have the best of both worlds in that erroneous DFT property calculations and limited experimental data compensate for each other.