Prediction of network level pavement treatment types using multi-classification machine learning algorithms

Abstract

The challenge in pavement management is to ensure that the correct treatment is applied to each segment of road at the appropriate time. Although pavement management systems assist with such predictions, pavement engineers further investigate and, ultimately, decide on the timing and type of treatment to be applied. While it is understandable that these might differ to some extent, large disparities between the two are coming under increased scrutiny. In order to assist with this challenge, this research develops a model for the prediction of pavement treatment types using multi-classification machine learning algorithms. The model attempts to predict what particular treatment type (reseal, overlay or rehabilitation) would be undertaken, based on available inventory and condition data. It also highlights the categories of data that were most influential in the prediction. The model was over 82% accurate in predicting the untreated segments and 80% accurate in predicting the treated segments.

KEYWORDS:

• Pavement management
• machine learning
• forward works planning
• treatment classification
• prediction
• pavement maintenance

Acknowledgements

The authors would like to acknowledge Fulton Hogan contractors for making the data available for use in this research and Lonrix Ltd for the use of their asset management system, JunoViewer, where the data was stored.

Disclosure statement

No potential conflict of interest was reported by the author(s).