COMBINATION OF ARTIFICIAL NEURAL NETWORKS AND GENETIC ALGORITHM - GAMMA TEST METHOD IN PREDICTION OF ROAD TRAFFIC NOISE
Abstract
This paper proposes an expert system based on Artificial Neural Networks (ANNs) to model road traffic noise. Feed-Forward
Neural Networks (FFNNs) that are trained with the Levenberg-Marquardt back-propagation algorithm were used. Models were
evaluated using mean squared error (MSE) and coefficient of determination (R2) as statistical performance parameters. In traffic
noise modelling, the noise level at a receptor position due to the source of traffic emission is modelled as a function of the traffic
conditions, road gradient, road dimensions, speed and height of buildings around the road. The curse of dimensionality problems
is caused by the large number of input variables in the ANN model. The Hybrid Genetic Algorithm-Gamma Test (GA-GT) as a
data pre-processing method for determining adequate model inputs was also evaluated. Genetic algorithms are frequently used
for the selection of input variables, and, therefore, reduce the total number of predictors. Through the hybrid model, six out of
twelve sets of predictor candidates were introduced as input variables in the ANN model. Comparing the results of the hybrid
model (ANN-GA-GT) with those of the ANN model indicates that the hybrid model has more advantages, such as improving
performance prediction, reducing the cost of future measurements and less computational and data storage requirements.
Consequently, the ANN-GA-GAMMA model is recommended as a proper method for predicting traffic noise level.