Treatment prediction using machine learning techniques. Sean Rainsford, Fulton Hogan


This presentation will describe efforts to utilise machine learning techniques, coupled with high speed condition and key asset data to construct a prediction model for future treatments. The approach adopted for this study utilized a list of completed treatments coupled with high speed data, surfacing records and maintenance records collected in the period leading up to the year in which the treatments were placed. The observed treatments (comprising of Seals, Asphalt or OGPA overlays and Rehabilitations) were used as a labelling set for the machine learning algorithm and several machine learning algorithms were trained to utilize available data to predict the probability of upcoming treatments in the next year. Several machine learning techniques are being explored, such as Nearest Neighbour (kNN), Decision Tree and Naïve Bayes models. At this stage of development, the Gradient Boosting algorithm and the Adaptive Boosting (ADAboost) model is providing the most promising results. The overall data set was split into a training set and a test set, with the test set being used solely as an out-of-sample set to evaluate the model accuracy. This work seems promising in terms of its ability to identify likely locations on the network where treatments will be needed in the near future, and similarly, to screen out segments where treatments will not be needed. The work is being carried out in conjunction with Lonrix Ltd, and the final model will be incorporated in the JunoViewer framework. It is hoped that the implemented models will be useful as a screening tool to prioritise field inspections, and also as an additional validation mechanism for deterioration model outputs.

Sean Rainsford works for Fulton Hogan as the Technical Asset Manager within the National Asset Management support team. Sean has been involved with data and pavement modelling for over 15 years, Sean was a part of the implementation of the dTIMS software into NZ. Sean’s passion is data and applicability to the real world. After driving many of the roads of New Zealand for over 20 years, and seeing the results of many forecasting outcomes, Sean has learnt that data is always the key!