The Particulate Matter Concentration Spatial Prediction using Interpolation Techniques with Machine Learning

Abstract

The air pollution problem have become the major global environmental problem. It also impacts to health, economic, traffic, and tourism of the nation. The air quality monitoring stations have been applied to measure the air quality factors in their surrounding area. However, the number of monitoring stations in developing countries may not be enough to cover the area. This paper proposes a framework to spatially predict the particulate matter concentration in the area without monitoring station. The proposed framework, called PAMS framework, consists of two components, which are 1) DUSTRY which is a particulate matter monitoring station to be deployed in a reference location, and 2) SPM which is a spatial prediction model to apply spatial interpolation technique and machine learning technique to provide the particulate matter concentration value in the area without monitoring station. This paper also explores the results from the variety of components in the PAMS. Two spatial interpolation techniques (i.e., IDW:Inverse Distance Weigh and Kriging) are compared. The evaluation results show that the the PAMS can spatially predict particulate matter concentration value with the average 10.16% error by using the Kriging technique with seven inputs for machine learning.