Comparison of Regression Model Concepts for Estimating Traffic Noise

Abstract

Traffic noise at two locations which are Rumuokoro and Rumuola in Port Harcourt city, Rivers state Nigeria was studied. The study was done for 3 days at each location. Variables such as atmospheric parameters and traffic density were measured along with the noise measurement. The atmospheric parameters measured were temperature, relative humidity and wind speed. Traffic density includes number of small cars and trucks per 20m radius. Three empirical model concepts were proposed, calibrated using multiple regression analysis and validated by cross validation and coefficient of determination (R2). The models are a linear model, a polynomial model and an exponential model. The coefficient of determination for the linear model ranged from 0.25 to 0.94 at Rumuokoro and 0.29 to 0.86 at Rumuola. The coefficient of correlation for the polynomial model ranged from 0.062 to 0.998 at Rumuokoro and 0.05 to 0.998 at Rumuola. The coefficient of correlation for the exponential model ranged from 0.28 to 0.92 at Rumuokoro and 0.45 to 0.89 at Rumuola. The exponential model was concluded to be the best model concept because of its performance in predicting noise levels using data from other days with moderately high and consistent coefficient of determination at both locations. However, if extrapolation is not to be considered, the polynomial model concept is very useful.