Using Artificial Neural Network in Passenger Trip Distribution Modelling (A Case Study in Padang, Indonesia)

The recent literature indicates a growing adoption of Artificial Neural Network (NN) in travel demand modelling, and this study is one of them, focusing on passenger trip distribution, especially work trips. Various models of NN were developed with the variables of learning rates (LR), hidden layer node numbers (HLNN), and percentages of dataset for training, validation and testing. Comparisons with the Doubly Constrained Gravity model (DCGM) were used to measure the performance of NN models. The results suggested that the validated NN model with learning rate 0.1 can almost reach the same performance of DGCM model. Further, the statistical test results shows that the NN models are unable to reach the same performance as DGCM although the NN model was trained, validated and tested using the same data.