Interpreting glass-forming rules across more than 78,000 compositions by common and system-specific analysis with SHAP-integrated XGBoost

AI editorPublished 2026.11.06Posted 2026.11.07 06:01 KST

A study published in npj Computational Materials analyzed glass-forming behavior by combining XGBoost classification with SHAP interpretation across 78,889 composition data points spanning conjugated, ionic, and metallic networks. The overall classification accuracy was 89.6%, and SHAP suggested a common frustration mechanism across different series and nonlinear critical behavior of compositional heterogeneity. [1]

1/ AI 활용 방법: SHAP를 결합한 XGBoost 분류
2/ 적용 과제: 78,889개 조성 데이터의 유리 형성 여부 분류와 지배 인자 해석
3/ 보고된 결과: 전체 데이터에서 89.6% 분류 정확도, 공통 frustration 메커니즘과 조성 이질성의 비선형 임계 거동 시사
4/ This study used 78,889 composition data points including conjugated, ionic, and metallic networks to classify glass-forming behavior and interpret which factors act system-specifically and commonly.
5/ The method was machine learning combined with SHAP and an XGBoost classifier.
6/ The study reported 89.6% classification accuracy on the full dataset, and SHAP analysis proposed a common frustration mechanism across multiple glass families and nonlinear critical behavior of compositional heterogeneity.
7/ From a materials research perspective, these results show that combining large-scale composition data with an interpretable classification model makes it possible to compare the governing factors of glass formation beyond simple prediction.
8/ In particular, handling different chemical series within one framework is useful for finding both universal rules and system-specific differences.
9/ However, the reported performance is limited to the corresponding dataset and model configuration.
10/ SHAP interpretation does not directly prove causality, and the interpretation of the critical behavior of compositional heterogeneity also requires additional experimental validation. [1]

How to apply this

Try predicting glass-forming behavior from large-scale composition data and checking interpretable feature contributions.

Organize the composition labels and features to create a training dataset.
Train an XGBoost classifier and check baseline performance with cross-validation.
Calculate global and individual feature contributions with SHAP and compare common factors with system-specific differences.
Check the reproducibility of prediction and interpretation on hold-out data or external data.

Data needed: You need composition data labeled with glass-forming behavior and composition-specific features or descriptors.
Checking the result: Use cross-validation while maintaining a train-validation split to confirm classification performance, and check whether the features highlighted by SHAP remain consistent in hold-out data.
What can go wrong: The most common failure is overfitting and causal misinterpretation, treating patterns that fit only the training data as general rules of glass formation.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.

웹 자료를 바탕으로 AI가 자동 요약한 글입니다. 연구의 세부 조건과 결과는 연결된 원문을 확인해주세요.

Interpreting glass-forming rules across more than 78,000 compositions by common and system-specific analysis with SHAP-integrated XGBoost

How to apply this

Related concepts and practice

Sources