An Improved Genetic Algorithm for Predicting Road Roughness Coefficient

In the realm of deep learning, this study introduces an enhanced genetic algorithm aimed at predicting pavement roughness utilizing data from the Long-Term Pavement Performance (LTPP) database. The initial step involves preprocessing the International Roughness Index (IRI) data through the application of box plot analysis and min-max standardization. This preprocessing effectively identifies and removes outliers, followed by normalization to ensure the reliability and accuracy of the dataset.Subsequently, the Genetic Algorithm-Back Propagation (GA-BP) neural network is trained by partitioning the dataset into training and testing subsets. This structured approach allows for a comprehensive evaluation of the model’s performance. The results yield a correlation coefficient $\mathrm{R} 2=0.84 \mathrm{R} 2=0.84$ and a root mean square error (RMSE) of 0.00270.0027, indicating a strong fitting effect and high predictive accuracy.These findings suggest that the proposed GA-BP model serves as an effective tool for evaluating road surface roughness, providing valuable insights for pavement management and maintenance strategies. The high level of prediction accuracy underscores the model’s potential application in the field of transportation engineering, facilitating data-driven decision-making for improving road quality.