Machine-Learning Modeling of Asphalt Crack Treatment Effectiveness

Abstract

Although many research projects have been completed to investigate asphalt crack treatment/repair effectiveness, all of them reported local investigation results with small sample sizes, incomplete time durations, and restricted types of sealants and/or treatment methods. Most importantly, machine-learning techniques have not been used to compare treatment methods and predict treatment effectiveness. In this study, all reported data regarding asphalt crack treatments were collected, and the large sample size data were comprehensively analyzed by statistical measures and machine-learning techniques to find the optimal crack treatment method. The variables including service time, environment temperature, type of sealant, treatment method, and traffic conditions were used as inputs, and the effectiveness of the asphalt crack treatment was selected as output information to compare the contribution of different variables for the treatment effectiveness and to provide prediction models. Chi-square test indicated that except for the variable “Traffic condition,” all other variables (Type of sealants, Treatment method, Service time, and Environmental temperature) had a significant relationship with the target variable (crack treatment effectiveness). The decision tree analysis results showed that the relative significance of the input variables to the target variable was ordered as Service time > Environmental temperature > Type of sealants > Treatment method > Traffic condition. The asphalt crack treatment effectiveness (%) can be predicted according to known variable values using the developed linear regression and artificial neural network (ANN) models. ANN model prediction has better performance than that of the linear regression model.