Core Concepts
ML framework integrating domain knowledge improves CFST axial capacity prediction accuracy.
Abstract
Introduces ML framework for accurate CFST bearing capacity prediction.
DKNN model enhances prediction accuracy by 50% compared to existing models.
Incorporates domain knowledge for robust predictions in noisy environments.
SHAP analysis identifies key parameters influencing axial load capacity.
Domain knowledge constraints improve model performance and stability.
Model ANNWT-5 outperforms other models in accuracy and reliability.
Robustness analysis shows ANNWT-5's resilience to noise.
MAPE values vary across different concrete and steel tube strength intervals.
Comparison with design codes shows the superiority of the ANNWT-5 model.
Stats
50% 이상의 예측 정확도 향상을 보이는 DKNN 모델.
SHAP 분석은 축하중량에 영향을 미치는 주요 매개변수를 식별합니다.
Quotes
"DKNN 모델은 기존 모델과 비교하여 예측 정확도를 50% 이상 향상시켰습니다."
"도메인 지식 제약 조건은 모델의 성능과 안정성을 향상시킵니다."