Development of an adaptive neuro fuzzy inference system based vehicular traffic noise prediction model

Noise pollution has been a global concern among the scientific community as it can cause long term and short-term adverse effects on human health. Vehicular traffic is one of the major causes of noise pollution. In the present work, an efficient methodology to predict the traffic noise level (L(eq)dBA) based upon vehicular traffic volume, percentage of heavy vehicles and average speed of vehicles has been proposed. To predict the noise level, adaptive neuro fuzzy inference system (ANFIS) has been developed and a detailed comparative analysis has been performed with conventional soft-computing techniques such as neural networks (NN), generalized linear model (GLM), random forests (RF), Decision Trees and Support Vector Machine (SVM). Implementation of ANFIS proof-of-concept model on testing data has resulted in higher accuracy for noise level prediction within 0.5 dBA and yielded significantly lower value of root mean square Error as compared to the conventional techniques. The results of current study signify the efficacy of the proposed method in prediction of traffic noise level and validate its suitability in planning mitigation measures for the new and existing roads. In order to analyse the performance of proposed technique, a case study of the highway locations near the city of Patiala in India has been presented.