An evolutionary computing approach for reducing bias in the dynamic modulus predictive models of hot mix asphalt

Abstract

Various published literature has indicated significant bias issues with dynamic modulus (E∗) predictive models for the hot mix asphalt. This study proposes Gene Expression Programming (GEP) based approach to reduce such bias. Bias analysis was carried out using database developed during national cooperative highway research program. GEP based predictive models for bias correction factors were developed and subsequently used in conjunction with E∗ predictive models to update E∗ values. Application of correction factors and subsequent statistical analysis showed improvement in the accuracy of predicted E∗ values. The extent of improvement (using sum of squared error) was dependent on model under consideration and ranged between 19 and 36%.