Capacity prediction for traffic circles: applicability of ANN

The traffic circle is a common solution which is widely used in urban and rural areas of Turkey. Although, most of the traffic circles are designed to be signal-controlled, some of them are still being used as unsignalised intersections in order to provide higher capacity and better performance particularly in rural areas. Regression analysis and gap acceptance-based models are the most used estimation methods for capacity prediction of unsignalised traffic circles. In this study, an artificial neural network model (ANN) was investigated as a new approach as ANN models have been successfully applied in various other traffic studies. The entry capacity was predicted by using exponential and multiple linear regressions, gap acceptance theory and the feed forward backpropagation algorithm type of ANN. The results were compared with well known models. The ANN model including the geometric parameters was found to be the most reliable estimator with 71.6% of proper predictions when the discrepancy percentages of the predicted versus observed entry flows were examined for the models. The multiple linear regression and gap acceptance models followed the ANN model with proper prediction proportions of 63.3 and 51.6%, respectively. On the other hand, models that included only circulating flow parameters were only found to be acceptable for limited geometric data conditions.