Potential cracking resistance indices for the SCB test utilising 100 mm diameter samples: Experimental investigation and machine learning analysis XGBoost-SHAP

This study aims to conduct an extensive and comprehensive investigation of range of potential cracking indexes available in the literature for the SCB test. In contrast to studies on 150 mm diameter SCB samples, this study focused on 100 mm diameter samples. This work is an effort to develop and accommodate SCB tests for specimens with 100 mm diameters prepared by either Marshall or gyratory compaction methods or cores from existing pavements. Four indices were compared and analysed, namely, the flexibility index (FI), cracking resistance index (CRI), cracking tolerance index (SCB-CTI), and rate-dependent cracking index (RDCI). The study used advanced analysis of machine learning (ML) integrating open-source library XGBoost and SHapley Additive exPlanations (SHAP) as support tools to analyse the sensitivity of the indexes, providing a clearer picture of the usefulness of each index. The study found that cracking indexes have good correlation to each other’s. Sensitivity analysis with the help of ML shows that CRI is the best choice among the investigated indexes. However, none of the indexes shows appropriate mechanical mechanism of the cracking resistance as lower air void samples provided lower indexes, indicating a room for improvement. Finally, the ML is found a robust technique to help scanning and evaluating the sensitivity analysis when it comes to proposing mechanical indexes for asphalt concrete testing.