Comparative Analysis of Traffic Congestion Prediction Models for Cellular Mobile Macrocells

Abstract

Traffic congestion prediction is a non-linear process that involves obtaining valuable information from a set of traffic data and regression or auto-regression linear models cannot be applied as they are limited in their ability to deal with such problems. However, Artificial Intelligent (AI) techniques have shown great ability to deal with non-linear problems and two of such techniques which have found application in traffic prediction are the Artificial Neural Networks (ANN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). In this work, Multiple Layer Perceptron Neural Network (MLP-NN), Radial Basis Function Neural Network (RBF-NN), Group Method of Data Handling (GMDH) and an Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are trained based on busy hour (BH) traffic measurement data taken from some GSM/GPRS sites in Abuja, Nigeria. The trained networks were then used to predict traffic congestion for some macrocells and their accuracy are compared using four statistical indices. The GMDH model on the average gave goodness of fit (R2), root mean square error (RMSE), standard deviation (?), and mean absolute error (µ) values of 99, 3.16, 3.53 and 2.32 % respectively. It was observed that GMDH model has the best fit in all cases and on the average predict better than ANFIS, MLP and RBF models. The GMDH model is found to offer improved prediction results in terms of increasing the R2 by 20% and reducing RMSE by 60% over ANFIS, the closest model to the GMDH in term of prediction accuracy.