Reliable and explainable machine learning for charge transfer/atomic structure relationships of hydrogenated nanodiamonds

A supervised learning model combined with genetic algorithm is proposed to predict charge transfer efficiency of nanodiamonds. The model is chosen among ten models whose parameters are modified by genetic algorithm with ten-fold cross-validation, ensuring the accuracy of model. Generalization abilit...

Full description

Saved in:
Bibliographic Details
Published in:Diamond and related materials 2024-04, Vol.144, p.110931, Article 110931
Main Authors: Wang, Peng, Ren, Jingli
Format: Article
Language:English
Subjects:
Electron affinity
Interpretability analysis
Ionization potential
Supervised learning
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A supervised learning model combined with genetic algorithm is proposed to predict charge transfer efficiency of nanodiamonds. The model is chosen among ten models whose parameters are modified by genetic algorithm with ten-fold cross-validation, ensuring the accuracy of model. Generalization ability and reliability are further verified by prediction error and consistency tests of twin nanodiamonds. A hydrogenated surface nanodiamond with low ionization potential and a clean nanodiamond with low electron affinity are designed based on particle swarm optimization. It is further found via SHAP analysis that electron affinity is inhibited by surfaces with a {111} surface below 30 % or a {100} surface above 80 %. Similarly, ionization potential is reduced when the hybrid ratios of sp1 and sp2 are separately greater than 0.501 % or lower than 5.690 %. [Display omitted]
ISSN:0925-9635
1879-0062
DOI:10.1016/j.diamond.2024.110931