SHORT TERM TRAFFIC PREDICTION IN SELECTED ROADS IN BENIN – CITY USING ARTIFICIAL NEURAL NETWORK

ABSTRACT

Short-term traffic speed prediction is a key component of proactive traffic control in the intelligent transportation systems. Traffic parameters including traffic flow, density and traffic speed are the dominate variables in short-term traffic prediction. The aim of this study is to predict short term traffic flow along Airport Road and Sapele Road in Benin City using artificial neural network. In this work, traffic study was conducted along Airport Road, from Oyaide junction to Irrhirrhi Junction, and Sapele road from JBS junction to Adesuwa junction using manual traffic count technique. Also, speed study was conducted on the vehicles that make use of the facility for the 15 minutes period. From these studies the time mean speed 𝑢𝑡 , the flow rate q, and the flow density k, were also determined. The different classes of vehicles which are passenger cars, light goods vehicles, heavy goods vehicles with 2 axles, and heavy goods vehicles with 3 or more axles and their respective time mean speeds, and the flow rate were used as the independent variables while the dependent variable used in the study was the flow density on both roads. The data were run on the machine learning App in Matlab 2018a by running the data on the ANN in the software. It was observed that for the road section along Sapele road, the number of best model occur when the number of hidden layer was 8 with an epoch of 773, and have the best validation performance of 8.2162 × 10−10. The goodness of fit of the training data and the testing data was 99.99%, which implies that the testing data perfectly interpreted the training data. And also along Airport Road it was also observed that the best model occurred when the hidden layer was 12 with an epoch of 301 iterations. The best validation performance occurred with a mean square error of 1.8282 𝑥 10−10. The goodness of fit was also 99.99%, which implies that 99.99% of the testing data can be interpreted by the training data and hence a perfect fit for the study. It was deduced that the ANN model is highly accurate in predicting traffic flow along both roads and has strong ability to generalise unseen data as shown by the low value of validation performance in both roads. The high goodness of fit suggests that, the model captured a significant portion of the variability in the traffic flow data and provides a highly accurate representation of the pattern from both roads