Improving Pavement Design by Estimating Resilient Modulus Using Experimental Data and Ensemble Methods

The resilient modulus of subgrade soil is closely connected to environmental variables, which influence soil characteristics and pavement behavior. Comprehensive understanding and interdisciplinary collaboration are crucial for effective pavement design and resilience to environmental changes. The resilient modulus (Mr) of subgrade soils plays a crucial role in structural design procedures for pavement design. However, due to regional disparities in soil characteristics and testing methodologies, determining Mr in the laboratory has become cumbersome. This research proposes a soft computing approach to predict Mr using models like Blending Ensemble (BD-ENSBL-(LR)), Support Vector Machine, Gradient Boosting Regression, Random Forest, and K-Nearest Neighbor Regression. Geological data from various locations in Ethiopia was used, with a 70/30 split for training and testing. The BD-ENSBL-(LR) model showed the best performance, with R² values of 0.99 (training) and 0.95 (testing) and low error metrics. The performance of built-in models is measured by three new index performance metrics of GB-R: a20-index of 100 and 97.62, Index of Agreement (IOA) of 1 and 0.99, and index of scatter (IOS) of 0.08 and 0.21 for training and testing, respectively, in addition to four common metrics. Feature importance analysis identified Nominal Maximum Axial Stress, clay content, and swell as key factors influencing Mr. The model's superiority was confirmed through statistical metrics and Taylor diagrams.