Predicting Macroscopic Parameters of Traffic Flow Using Deep Learning Methods

In recent decades, with the development of transportation infrastructure and the increase in population in urban areas, traffic congestion has become one of the main challenges of urban life. The excessive growth of cities, the increase in the number of cars and motorcycles, and the lack of proper traffic management have brought about numerous problems. These problems include air pollution, excessive fuel consumption, increased accidents, loss of time, and increased stress in citizens. To address this challenge, the use of intelligent transportation systems has been proposed as an effective solution. In this study, network search has been used to optimize hyperparameters for both MLP and CNN-LSTM models. Early stopping techniques have been used to prevent overfitting to ensure that the models generalize well to unseen data. The performance of the models is evaluated using the root mean square error (RMSE) and mean absolute error (MAE), which are standard metrics for assessing prediction accuracy. In this study, two models, a multilayer perceptron network and a convolutional neural network-long-short-term memory, are used, and here the architecture used in this paper is examined in more detail. The results obtained show that the CNN-LSTM model has much better prediction ability than other models, especially in traffic prediction under non-repetitive conditions. The gated recurrent unit, like the long short-term memory, does not have an output gate. The gated recurrent unit has fewer parameters, so it is more computationally efficient and requires less data for generalization than the long short-term memory.