Accurate prediction of laboratory permeability of hot mix asphalt using machine learning techniques

Estimation of permeability of Hot Mix Asphalt (HMA) is a critical step towards ensuring moisture resistant mixes and preventing premature failure in flexible pavements. The objective of this paper is to present an approach of using advanced machine learning techniques for accurate prediction of permeability. The predictors are commonly obtained parameters such as percent passing different sieve sizes and voids—either from saturated surface dry method or from vacuum seal method. An additional parameter, porosity, which could be obtained from the testing of bulk specific gravity by the vacuum seal methods is also used. Instead of predicting permeability values, an approach of classification, according to high and low permeability is presented using the Support Vector Machine and the Naïve Bayes methods. Both approaches showed excellent accuracy of prediction. The Gaussian Process Recognition (GPR) method was also found to be very successful in predicting permeability values with high accuracy. Use of classification and appropriate machine learning techniques are recommended for research in pavement engineering.