Use of data mining techniques to explain the primary factors influencing water sensitivity of asphalt mixtures

Abstract

The water sensitivity of asphalt mixtures affects the durability of the pavements, and it depends on several parameters related to its composition (aggregates and binder) and the production and application processes. One of the main parameters used in the European Standards to measure the water sensitivity of asphalt mixtures is the indirect tensile strength ratio (ITSR). Therefore, this work aims to obtain a predictive model of ITSR of asphalt mixtures using several parameters that affect water sensitivity and assess their relative importance. The database used to develop the model comprises thirteen parameters collected from one hundred sixty different asphalt mixtures. Data Mining techniques were applied to process the data using Multiple Regression, Artificial Neural Networks, and Support Vector Machines (SVM). The different metrics analysed showed that SVM is the best predictive model of the ITSR (mean absolute deviation of 0.116, root mean square error of 0.150 and Pearson correlation coefficient of 0.667). The application of a sensitivity analysis indicates that the binder content is the parameter that most influences the water sensitivity of asphalt mixtures (26%). However, this property depends simultaneously on other factors such as the characteristics of the coarse and fine aggregates (24.9%), asphalt binder characteristics (19.3%) and the use of additives (10%).