Generalized LIME-based feature ranking for California bearing ratio prediction using SHAP and LIME–BORDA approaches

Abdullah Hulusi Kökçam

doi:10.1016/j.aei.2026.105129

Generalized LIME-based feature ranking for California bearing ratio prediction using SHAP and LIME–BORDA approaches

2026-08-14

0

PRE

AI

A

Abdullah Hulusi Kökçam

U

Uğur Dağdeviren

Т

Талас Фикрет Курназ

A

Alparslan Serhat Demir

C

Caner Erden *

DOI：10.1016/j.aei.2026.105129

Original

Original request for help

Save

Abstract

En 中文

• Twelve machine learning regression models were benchmarked for CBR prediction. • Random Forest provided the highest generalization performance with a test R2 of 0.832. • A novel LIME–BORDA framework generalized local explainability via weighted rank aggregation. • Generalized LIME–BORDA results showed a high correlation with global SHAP importance. • Fines content and maximum dry density were identified as the primary governing variables.

Keywords:

California bearing ratio

Explainable AI

SHAP

LIME

BORDA

Generalized LIME-based feature ranking for California bearing ratio prediction using SHAP and LIME–BORDA approaches

Abstract

Journal

Organization