An approach to identifying families in roughness progression through design group classification and MML inference.

The use of family modelling to divide pavement networks into groups with similar within group deterioration is quite extensive. This popularity is due to the combination of relatively simple models and engineering justification for the approach. It seems reasonable, for instance, that different pavement types within a network may deteriorate at different rates, even with the same climate, traffic pavement depth, age etc. The advantage of dividing the network into families is that the model can implicitly include the variables that define the families without having to include them in the deterioration function. Further, complex statistical issues of multi co-linearity can be effectively bypassed. The disadvantage in family modelling is the difficulty in correctly identifying the appropriate families and subsequent high risk in under/overfitting. This paper presents a new approach to identify families with two components. Firstly, the network is classified into design groups to reduce the number of possible families making searching simpler and secondly, a minimum message length (MML) criteria to quantitatively select the appropriate family model. We perform simulated comparisons comparing common criteria to select family models and conclude that MML is the preferred criterion. (a) For the covering entry of this conference, please see ITRD abstract no. E217099.