Table 1 Summary of the results from case studies of modeling the experimental band gap (Eg), lattice thermal conductivity(κL), and elastics of zeolite (log(K))

From: A strategy to apply machine learning to small datasets in materials science

Property E g κ L log(K)
Data volume 108 93 102
Scaled error (%) 6.2 4.1 6.1
Scaled error before (%) 9.3 6.2 13
DoF 9 5 -a
DoF before 12 7 -a
Source of property Experiment Experiment DFT
Source of CEP DFT Empirical model Force field calculation
Regression method Kernel ridge Kernel ridge Gradient boosting
Feature selection Stepwise forward search Stepwise forward search a
  1. aFollowing the same approach used in ref. 31 no feature selection was performed for log(K)