Moisture damage prediction of polymer modified asphalt binder using Support Vector Regression

This paper reports the use of Support Vector Regression (SVR) for modeling moisture damage in asphalt binder. This work is based on the nano-scale testing reported earlier by the authors where conventional approach were applied in describing the moisture damage phenomena. The testing on dry and wet asphalt binders were conducted under an Atomic Force Microscopy (AFM). Parameters were varied as: percentages of polymer (SB), percentage of anti-stripping agents, AFM tips K value and AFM tip types based on different chemical functional of asphalt binder for the testing purpose. The combination of these input features were varied through application of hill climbing approach to select the best parameters set that can describe the damage in asphalt binder accurately. The experimental result suggests that a very high accuracy (for dry asphalt samples NRMSE = 4.109 and CC = 0.908; for wet asphalt samples NRMSE = 2.881 and CC = 0.912) in the prediction of adhesion force value for asphalt binder can be achieved using SVR.