| Sumario: | Nation-wide Annual Average Daily Traffic(AADT)data on NFAS roads across the country are destroyed. Two machine learning methods, the Artificial Neural Network and Random Forest demonstrate a substantial increase in the accuracy of estimating AADT according to five scales that is MSE, RSQ, RMse, MAE and MAPE, using a Spatial Autoregressive Model as a benchmark. An estimated AADT of 87 variables in the area of central, adjacent traffic, population, jobs, land-use diversity, density of road networks, urban design, destination access, etc. is focused on data mining from three aspects, i.e. on road and off-road, network centrality and neighbouring influences. The variable collection for estimates is promoted by aggregation of data by different buffer sizes and linearity and singletonity statistical analysis. The interplay between the variables, variable measurements of significance are extensively explored when applying machine-learning approaches not only the estimation output but also the relationship between and variable and AADT.
|
|---|