Thermal features prediction in asphalt pavements using ANFIS-based regression

This paper aims to develop predictive models for the thermal properties of laboratory-prepared hot mix asphalt (HMA) specimens using the Adaptive Neuro-Fuzzy Inference System (ANFIS). The thermal properties investigated include thermal conductivity, thermal diffusivity, and specific heat. Thirty specimens were prepared by varying the mixture’s nominal maximum aggregate size and gradation coarseness, using a single asphalt binder. The transient plane source method was employed to determine the thermal properties, and volumetric parameters such as air void volume and effective binder volume were calculated. Surface characteristics, including friction and texture, were also considered. A total of 150 data points were generated through three aggregate sizes, two gradation levels, five replicates, and five measurement locations to ensure accuracy and repeatability. The model development process was structured in three phases: (1) feature reduction to detect multicollinearity, (2) outlier detection and removal, and (3) application of the ANFIS regression model. Model performance was evaluated using mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R2). Results indicated that outliers significantly increased MSE and reduced prediction accuracy, with RMSE improving from 0.4752 to 0.1765 after outlier removal for thermal conductivity. R2 scores for thermal diffusivity and specific heat improved from 0.55 and 0.72 to 0.866 and 0.91, respectively, after addressing multicollinearity and outliers. The study demonstrates that the proposed ANFIS model effectively predicts the thermal properties of HMA specimens and highlights the potential for even greater predictive accuracy by integrating other advanced regression methods.