Interpretable machine learning of formation energies and band gaps in hybrid perovskites through SHAP analysis and SISSO modelling
Abstract
Accurate and interpretable property prediction is essential for exploring hybrid organic-inorganic perovskites. Here, we combine machine-learning screening with SHAP analysis and SISSO descriptor discovery. Using 4,456 compounds, optimized XGBoost models achieved test-set R2 values of 0.99 for formation energy and 0.93 for band gap and were applied to over 180,000 candidates. Comparison across approximately 17,000 common compounds revealed broadly overlapping distributions from XGBoost and CGCNN but substantially lower predictions from UPET, suggesting potential limitations in its transferability to hybrid organic-inorganic perovskites. SHAP analysis and correlation-based pruning retained eight primary features for each property, from which SISSO derived expressions comprising four nonlinear descriptors. These expressions couple molecular and inorganic properties for formation energy and emphasize relative atomic sizes and B-site characteristics for band gap. This framework supports efficient candidate prioritization and interpretable analysis of learned structures-property relationships in hybrid perovskites.
Related
- Native point defects in rock-salt and wurtzite MnSe: formation energies across the growth window and spin splitting in the A-type order
- Lead-free piezoelectric perovskites for arterial-pulse e-skin: from configurational complexity to equivariant machine-learning potentials
- Deep Learning for Longitudinal Asymmetry in Heavy-Ion Collisions at RHIC Energies
- Machine learning-enabled prediction of dynamic vertical triplet energies underscores the role of vibrational anharmonicity in energy transfer
- Harnessing Machine Learning and Composite Quantum Chemical Methods for Computing Accurate Bond Dissociation Energies of PFAS
BibTeX
@misc{xia2026interpretable,
title = {{Interpretable machine learning of formation energies and band gaps in hybrid perovskites through SHAP analysis and SISSO modelling}},
author = {Christopher Xia and Hao Zhang and Qingyuan Jin},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.09944},
eprint = {2611.09944},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.09944},
}
RIS
TY - UNPB
TI - Interpretable machine learning of formation energies and band gaps in hybrid perovskites through SHAP analysis and SISSO modelling
AU - Christopher Xia
AU - Hao Zhang
AU - Qingyuan Jin
PY - 2026
DA - 2026/11/01
PB - arXiv
DO - 10.48550/arxiv.2611.09944
UR - https://arxiv.org/abs/2611.09944
AB - Accurate and interpretable property prediction is essential for exploring hybrid organic-inorganic perovskites. Here, we combine machine-learning screening with SHAP analysis and SISSO descriptor discovery. Using 4,456 compounds, optimized XGBoost models achieved test-set R^2 values of 0.99 for formation energy and 0.93 for band gap and were applied to over 180,000 candidates. Comparison across approximately 17,000 common compounds revealed broadly overlapping distributions from XGBoost and CGCNN but substantially lower predictions from UPET, suggesting potential limitations in its transferability to hybrid organic-inorganic perovskites. SHAP analysis and correlation-based pruning retained eight primary features for each property, from which SISSO derived expressions comprising four nonlinear descriptors. These expressions couple molecular and inorganic properties for formation energy and emphasize relative atomic sizes and B-site characteristics for band gap. This framework supports efficient candidate prioritization and interpretable analysis of learned structures-property relationships in hybrid perovskites.
ER -