Neural networks model to estimate traffic capacity for weaving segments

The impact of weaving vehicles on the capacity of freeway segments is uncertain due to the complexity in operation. The Highway Capacity Manual 2000 provides values for capacity on various weaving segments (Exhibit 24-8) based on sets of conditions (configuration, speed, length, volume ratio, and number of lanes). However, to find capacity for a given set of conditions, an iterative process should be carried out using a properly programmed spreadsheet. This paper suggests alternative and convenient procedure for estimating capacity on weaving segments. Two capacity prediction models are developed using regression and neural networks. Although, linear regression technique showed satisfactory results, neural network technique outscored linear regression in the prediction performance, and generalization ability. The trained neural network architecture represented by weight and bias values for each layer is simply used to predict capacity for weaving segments under new conditions.