The use of data mining techniques for road maintenance planning

Classic analysis techniques are inadequate when new correlations and dependences among data, which become more numerous and complex every day, have to be discovered. Data mining techniques allow us to classify sections with different common characteristics (cluster). These two properties (homogeneity of characteristics within each class, and unlikeness of characteristics among the classes) are fundamental for the analysis to be carried out. A methodology is proposed to evaluate conditions that compromise road efficiency, as a support for choosing the right interventions in order to improve the road system. In this study, it was decided to use Bayes nets because they can work even with incomplete records. In fact, databases for road managing are often incomplete. The first step is Bayes net training, i.e. showing the net some examples with specifications about the class they belong, so the net could learn as much as necessary to distinguish samples belonging to different classes. The next step consists of a test to verify if the net has learnt correctly. The net is used as a classifier, submitting samples never shown before, whose class is not known. This system can support efficient planning of road maintenance activities, organized in an information system, expressly created to satisfy these kinds of needs. For the covering abstract see ITRD E144473.