PERFORMANCE OF DEEP NEURAL NETWORKS IN THE ANALYSIS OF VEHICLE TRAFFIC VOLUME

Abstract

The major problem of vehicle traffic congestions is the increased time wasted in the queues and the resultant high cost of resources used during the same period. Therefore, this research seeks to evaluate the viability of Deep Neural Networks in the performance analysis of vehicle traffic volume. This will assist in effective and efficient traffic monitoring, travel-time forecasting and traffic management. Deep Neural Networks (DNN) offer an optimal option for alleviating the problem of traffic congestion. Although Artificial Neural Networks (ANN) usually encounter setbacks such as local optimum thereby resulting in short term forecasting this can be effectively overcome by using an appropriate training algorithm with correctly configured parameters for the kind of data under consideration. The data is divided into samples for training, validation and testing, after which the overall performance is evaluated using the Mean Squared Error (MSE). The results obtained will help in the evaluation of the practicability of using DNNs in analyzing vehicle traffic flow. Eventually, this can be leveraged for time-forecasting of traffic conditions and also mitigate traffic build-up.