Particulate Matter 2.5 Machine Learning Estimation System Based on Internet of Things Architecture

This study designs a mobile air pollution sensing system to monitor the concentration of particulate matter 2.5 in the metropolitan area based on the Internet of Things. The measurement system can improve the weakness that sparse data of government-managed fixed monitoring stations, and can more accurately describe the pollution levels of various areas of the metropolitan. The estimation system uses four regression models which contain Decision Tree (DT), Random Forest (RF), Multilayer Perceptron (MLP) and Radial Basis Function Neural Network (RBFNN). The results show that RF is better than other regression models in training set and testing set. In order to verify the generalization ability of the learning model, we selects three days for field verifications, on 2019/02/15, 2019/02/28 and 2019/03/01. The predictions can be visualized through the web application, and the map form allows the user intuitively understand the distribution of the contaminated area.