Improving Aggregate Abrasion Resistance Prediction via Micro-Deval Test Using Ensemble Machine Learning Techniques

Abstract

Aggregate is the most extracted material from the world's mines and widely used in civil and construction projects. The Micro-Deval abrasion test (MD) is one of the most important tests that provides characteristics of crushed aggregates that show their resistance against mechanical abrasive factors such as repeated impact loading. The impact of various factors on abrasive resistance properties of aggregates has led researchers to seek correlations, often focusing on limited data samples, leading to reduced accuracy. This study employs machine learning (ML) methods to predict MD abrasion values, considering diverse aggregate properties. Various ensemble ML methods were applied, revealing the exceptional performance of the stacking model, which achieved an R² score of 0.95 in predicting aggregate abrasion resistance. The feature importance analysis highlights the influence of factors such as Magnesium Sulfate Soundness (MSS), Water Absorption (ABS), and Los Angeles Abrasion (LAA) on aggregate abrasion values, suggesting that the use of multiple test methods could yield a more dependable assessment of aggregate durability.