Comprehensible artificial intelligence-based models for raveling of porous asphalt

Among a large number of existing artificial intelligence (AI)-based techniques, artificial neural network (ANN) is one of the most widely used techniques, which has been applied to many road engineering problems. They are especially useful as predictive models because they do not require prior knowledge of the input data distribution and have shown high prediction performance. However, despite their high degree of accuracy, these AI models are difficult to interpret. For many problems, it is desirable to extract knowledge in a comprehensible manner so that the users can gain a better understanding of the solution. The traditional if-then rules are usually the most understandable ones. In the Netherlands, the surface damage of porous asphalt, being raveling, draws a lot of attention because porous asphalt (PA) has been applied to more than 70% of Dutch highways. A better understanding of this problem can lead to lower maintenance cost. Therefore, this paper applies two comprehensive modeling techniques to the problem of raveling. The first technique is rule extraction from function approximating neural networks (REFANN), which extract rules from ANN. The second one is regression trees. These modeling techniques were applied to the data of 72 PA sections obtained from SHRP-NL.