Neural networks for performance prediction on unsealed roads

In this project a large database on the performance of unsealed roads was re-analysed using neural networks to determine if any improvements to the current prediction models could be made. The data analysed includes overall performance and gravel loss for various materials used for unsealed roads in South Africa. The data was analysed with various forward-feed networks and the results compared with those already derived by statistical analysis. The results are promising, resulting in higher correlations and more accurate predictions, because the networks can approximate very complex functions. However, the incorporation of these networks into existing unsealed road management systems is still being investigated.