01661434

Component

This study compared pavement marking retroreflectivity prediction using artificial neural networks (ANN) and linear regression (Ordinary Least Squares, OLS). Using retroreflectivity data collected from the newly applied pavement markings along different roadways throughout the state of Tennessee, the study quantified the degree of prediction accuracy from the two modeling approaches. Comparing the observed and predicted retroreflectivity values in terms of coefficient of determination (R-Squared), the ANN was found to be more accurate and precise predictor compared to linear model, in some instances by a big margin. Combining white markings data together (paints and thermoplastic), ANN prediction accuracy was recorded at 77% compared to 47% for OLS, a better performance by 64% in favor of ANN over OLS. For yellow markings combined, ANN outperformed OLS by 40%. When marking colors were breakdown in terms of paints and thermoplastics, ANN outperformed OLS by 24%, 67%, 300% and 63% for white paints, yellow paints, white thermoplastic and yellow thermoplastic respectively. Though performed better than OLS, the study found that only well-structured and organized ANN network with optimized number of neurons and epochs perform better than a traditional statistical technique such as OLS in the prediction. For performance prediction, ANN does not require any predefined assumption during computation and it is very easy for developing an algorithm unlike linear models where statistical concepts and interpretation results is required prior to analysis. Though having high predictive ability and flexibility of modeling, ANN has a weakness that the internal operation cannot be accessed.

This paper was sponsored by TRB committee AHD55 Standing Committee on Signing and Marking Materials. Pavement Marking Retroreflectivity Service Life: Comparing Regression and Artificial Neural Network Predictions: This is an alternate title.

18-04181

English

13p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; References; Tables

Deterioration; Linear regression analysis; Mathematical prediction; Neural networks; Retroreflectivity; Road markings; Service life

Highways; Materials; Pavements

Transportation Research Board Annual Meeting 2018 Paper #18-04181

TRIS, TRB, ATRI

Jan 8 2018 11:02AM