Integrated FFT and XGBoost framework to predict pavement skid resistance using automatic 3D texture measurement

In this study, an integrated fast Fourier transform (FFT) and an extreme gradient boosting (XGBoost) framework were developed to predict the pavement skid resistance using automatic 3D texture measurement. The 3D pavement surface data was segmented into macro and micro-texture using the FFT and then characterized by 23 different parameters. Subsequently, the XGBoost algorithm was applied to develop the pavement friction pre-diction model and to assess the attributes of the proposed parameters. The results indicated that the Skewness (R-sk), Mean Profile Depth (MPD), and Power Spectral Density (PSD) were most applicable to characterize the pavement micro-texture in terms of the skid resistance. Further, the XGBoost model was found to achieve excellent prediction accuracy with an R-2 value of 0.88, outperforming the multiple linear model (R-2 = 0.71), decision tree model (R-2 = 0.76), and the random forest model (R-2 = 0.86) that were investigated in this study.

Automated summary

This study developed an integrated FFT and XGBoost framework to predict pavement skid resistance with high accuracy, with Skewness, Mean Profile Depth, and Power Spectral Density being identified as key parameters for characterizing pavement micro-texture. The XGBoost model outperformed other models with an R-2 value of 0.88.