Road roughness modelling with clustered data using ANN approach

It is a well established fact that the roughness is manifested as an effect of different individual pavement deterioration parameters. Several studies have been oriented in the direction of establishing the models capable of predicting the roughness. However, it was felt essential to develop a model explaining the dynamics of different pavement deterioration parameters on the roughness. In view of very limited studies reported in this direction, the present study is carried out, first by grouping available data into homogeneous clusters and then model them using Feed Forward Back Propagation Artificial Neural Network algorithm. K- Means partitional clustering algorithm has been adopted for clustering the data. A new mathematical algorithm has been proposed and used for optimizing the number of clusters, which is further verified with the available standard validity indices. The present modeling attempt has indicated strong correlation between the road roughness and the deterioration parameters viz cracking, raveling, potholes, patching and rutting. The models developed for all the clusters have shown decent statistical acceptability.