講演情報

[AI-04]International Friction Index (IFI) Prediction with machine learning for Pavement Management System

*Ponlathep Lertworawanich1, Songyasin Chonpatathip2 (1. Department of Highways (Thailand), 2. Ministry of Transport (Thailand))

キーワード:

International Friction Index (IFI)、Machine Learning、Pavement Management System

Pavement friction is an important parameter that affects safety of roadways. Low friction roadways are more likely to accidents than high friction roadways. There are numerous studies that examine the relationship friction and accident rates on highway networks. It was found that the rate of wet crashes increased when pavement friction was low. On the contrary when a high skid resistance is available, the rate of wet crashes is reduced. Pavement friction data are usually collected using complicated equipment which require high operating budgets and time-consuming to cover all roadways in the Thailand highway networks. However, friction deterioration models with sufficient prediction accuracy are not yet available. This study develops the International Friction Index (IFI) prediction model based on machine learning algorithms such as neural networks, decision tree models, random forest models, and linear regression models. There are two categorical (surface and aggregate type) and nine numerical predictor variables (such as MPD, AADT, number of lanes, and truck traffic in different sizes). The numerical variables are processed with the Principal Component Analysis (PCA) with the 95% variance threshold to reduce the number of variables in the prediction model. It is found that the decision tree and random forest models can provide good estimates of IFI with R2 of 0.70 where the multiple linear regression and artificial neural network can give R2 of 0.15 and 0.47. In future, the proposed IFI prediction model will be incorporated into the Thailand Pavement Management System.