Authors: Xiaozhe Yin*, University of Southern California, Meredith Franklin, University of Southern California, Masoud Fallah Shorshain, University of Southern California, Yao-Yi Chiang, University of Southern California, Scott Fruin, University of Southern California, Rob Scot McConnell, University of Southern California
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems, Transportation Geography
Keywords: traffic noise, mobile data collection, spatial modeling, machine learning
Session Type: Paper
Start / End Time: 3:20 PM / 4:35 PM
Room: Virtual Track 2
Presentation File: No File Uploaded
With growing urbanization and expansion of urban transportation needs, traffic noise has become a dominant environmental issue, and therefore, a particular interesting research problem. Traditional fixed data collection method to measure traffic noise cannot capture its spatial heterogeneous characteristics. Therefore, we conducted a novel mobile data collect method which is capable of identifying small-scale spatial patterns for more polluted areas. The collected data was further compared with traffic noise from TNM2.5. The results indicate that TNM2.5 overestimate more than 15dB in north Long Beach area. In order to understand the contributions of different features and predict traffic noise, we adopted a deep learning technique which consist of geographical, weather, traffic, and other built environment data as well as the spatial pattern of traffic noise. The result of the model will be compared with other baseline machine learning models, including XGBoost, Random Forest, and Multilayer Perceptrons (MLPs) to evaluate its performance.