New article on road maintenance out now

What was the starting point for the work?
– The aim of the study was to develop a simple and clear methodology to help the municipality make good decisions on road maintenance. The aim was to use the maintenance budget in a smart way and keep the streets in good condition. The study was conducted in Skellefteå municipality and is about how to use machine learning (random forest (RF), neural networks (NN) and linear regression (LR)) to predict street conditions based on subjective assessment data. The study helps to predict street kicks, allowing municipalities to plan maintenance well in advance and use the budget in a smarter way.

What are the main conclusions?
– Four key conclusions are these:

1. The RF model performed best in predicting street behavior for both residential and non-residential streets, compared to NN and LR models.
2. Road age is the most important factor in all models, but other variables such as damage and traffic are also important depending on the street type.For example, for residential streets, age and road damage were most useful, while for non-residential streets, age and traffic were most useful.
3. Simple models like LR can work well, especially for smaller municipalities with limited budgets – the models help to prioritize maintenance and use resources smartly.
4. Increased data collection is recommended: As the study is limited to one municipality and only two street inventories (2014 and 2018), more inventories of the street network over time are suggested to better understand how the models perform. In addition, similar studies in other municipalities would improve the usability of the models at municipal level.

Who do you think should read the article, and what do you hope they take away?
– People working in road maintenance and road management at the municipal level, especially those interested in using machine learning to improve decision-making and resource utilization. Also researchers and engineers focusing on road maintenance, road management and machine learning applications in urban planning and infrastructure. This study can also help to prioritize the selection of street maintenance objects to optimize the condition of the street network for a given budget.

Amjad Afridi co-authored the article with Sigurdur Erlingsson, Leif Sjögren and Cristofer Englund. Read the paper here: Predicting Pavement Condition Index Using an ML Approach for a Municipal Street Network