Predicting dynamic modulus for bituminous concrete using support vector machine

Absolute value of complex modulus is dynamic modulus. Complex modulus of bituminous mixes is a key parameter used in mechanistic design approach of flexible pavements. Performance of bituminous mix including fatigue cracking and rutting in flexible pavement also depends upon its complex modulus. Therefore, in order to better understand behavior and performance of flexible pavements, assessment of complex modulus of bituminous mixes is inevitable. It is well understood that complex modulus of bituminous mixtures are function of temperatures and loading frequencies. Laboratory investigation of complex modulus requires sophisticated equipments, tedious time consuming experimentation and intricate computation work, which encourages modeling techniques. In the recent past researchers have developed models for predicting dynamic modulus of bituminous mixes using finite element (FE), artificial neural network (ANN) and regression-based techniques. However, limited models have been developed for Indian temperature condition and bitumen grades. In the present study, dynamic modulus of bituminous concrete (BC) using artificial neural network (ANN) and support vector regression (SVR) based modeling approach has been developed. The dataset consisting of 80 measurements derived from an earlier study was used. As the choice of the methods affects the performance of a machine learning algorithms and may influence the accuracy of the final model, thus three methods for model testing were used to evaluate the effectiveness of SVR modeling technique in predicting dynamic modulus of bituminous concrete. Analysis of results indicates a comparable performance by (i) 10-fold cross validation and (ii) LOOV method in terms of correlation coefficient and means absolute error values. In random division method choice of a random number affect the model performance. This study may be helpful in fair estimation of the dynamic modulus value for bituminous concrete at different temperature and load frequencies prevailing in India.