Machine learning-based Air Temperature Prediction for

Asphalt Pavement Binder Classification

Project Overview

Performance-based grading of asphalt binders used in the construction of flexible pavements is done on the basis of the 7-day maximum and 1-day minimum pavement surface temperatures. These pavement surface temperatures are usually predicted using the air temperature, 2 m above the pavement surface. The accuracy of the pavement surface temperatures depends on the accuracy of the air temperatures. Real-time air temperature data is available from the local weather stations; however, in the context of pavement design, air temperatures over the service life of the pavement are required. Therefore, a study of air temperatures in the future based on the historical records is required.

The objectives of this study are to:

1) Develop gene expression programming (GEP)-based analytical equations for the prediction of air temperature as a function of local weather characteristics as well as geographical coordinates.

2) Develop gradient-boosting-based machine learning models for the prediction of air temperature and to understand the effect of clustering data into smaller subsets on the prediction accuracy.

Gene Expression Programming based Analytical Models for the Prediction of Air Temperature

The methodology implemented for developing the GEP-based models is shown in the flow chart, and the developed GEP models for the prediction of 7-day maximum and 1-day minimum air temperatures are listed below. The developed models can be used for predicting the air temperatures at different reliability levels.

Methodology for Optimizing the Tree Depth

The predictability of the model improves as the tree depth of the GEP model is increased. However, increasing the tree depth comes at the cost of increasing the complexity of the analytical model, which makes it difficult for hand-calculations. Moreover, there is a risk of overfitting. Therefore, it is preferable to have GEP models with smaller tree depths, yet with a reasonable prediction accuracy. In this study, an optimal tree is selected by plotting the counter trends of coefficient of determination and the standard deviation in the prediction error. Tree depths beyond which the changes in the coefficient of determination were only incremental were chosen as the optimal tree depths.

Refer the following article for more details

Padala, S. K., Swamy, A.K., Bhattacharjee, B. (2025). Air temperature prediction models for pavement based on gene expression programming approach. Journal of Transportation Engineering, Part B : Pavements (ASCE). 151(1). https://doi.org/10.1061/JPEODX.PVENG-1496

Assessing the Effect of Clustering on the Prediction Accuracy

Gradient-boosting-based machine learning models for prediction of air temperatures are developed using the above data. The effect of clustering the data into smaller subsets on the prediction accuracy is studied. The entire country was classified into several geographical clusters using (i) k-means clustering (5-15 clusters) algorithm, (ii) empirical clustering, and (iii) political state boundaries. The clustered regions as per the various clustering schemes are shown in the figure in different colours.

The sum of squared error between the predicted and actual values is compared for different clustering schemes on various datasets.

Refer the following article for more details

Padala, S. K., Kumar, S., Swamy, A.K., Rao, R.K. (2024). Air temperature prediction models for pavement design: a gradient boosting-based approach. International Journal of Pavement Engineering (Taylor & Francis). 25(1). https://doi.org/10.1080/10298436.2024.2381658

Page updated

Google Sites

Report abuse