Development of a Pavement Condition Prediction Model Using Machine Learning
This project was part of the Bachelor’s Final Year Project that I supervised. The students involved in this research were Kash, Mash, and Yifan. Together, we developed a Pavement Condition Prediction Model that leverages machine learning techniques. The focus was on creating a system to predict road roughness using historical pavement data, traffic data, and weather information.
Project Overview
The primary goal of this research was to utilize machine learning models to predict pavement conditions by processing crowdsourced smartphone data. Our methodology involved data collection, pre-processing, and the application of machine learning algorithms like MiniBatch Linear Regression and Random Forest Regression.
We specifically focused on state roads within the City of Monash, collecting roughness data from VicRoads, climate data from local weather stations, and traffic data to feed into the models. The pavement prediction process used this combination of data to train and test the machine learning algorithms.
Key Insights
- Random Forest Regression outperformed the MiniBatch Linear Regression, with better predictive accuracy and robustness against outliers.
- Data pre-processing played a crucial role, including filtering for features like rainfall, temperature, and traffic volume.
- Although the Random Forest model showed promise, further improvements are necessary to enhance accuracy for real-world applications.
Future Work
We recommend expanding the spatial coverage of the data and integrating more features, such as pavement rutting and lane-specific data, to improve the model’s performance. Additionally, increasing the granularity of time intervals for data collection could help achieve better predictions over longer periods.
Our findings show the potential of machine learning to transform how we predict road conditions, offering a more cost-effective and timely approach to pavement maintenance and planning.