A Comparative Study of Deep Learning Model and Simple Prediction Charts in Construction Noise Prediction

Abstract

Construction noise monitoring is crucial to assess the impacts of construction noise on the workers and surroundings. However, the existing noise prediction methods are time-consuming in which required laborious work for the computation of noise levels. This study aims to assess the accuracy and reliability of deep learning model (DL) that adopted stochastic modelling and artificial neural network (ANN) in construction noise prediction. The artificial neural network was trained with the output of stochastic modelling. The outcome of noise level prediction using simple prediction chart (SPC) and DL model was discussed and compared to 3 case studies. The case studies were conducted at construction sites located in Semenyih, Selangor, Malaysia. The results of DL model showed high accuracy of predicted noise levels along with an absolute difference of less than 2.3 dBA. Besides, the predicted noise levels are reliable as the R-squared value is higher than 0.992. On that account, DL model is proved to be reliable and accurate in noise level prediction and it has the potential to be utilized as a managerial tool to monitor construction noise more effectively.