Augmented sensors for particulate matter concentration prediction using supervised learning models

Abstract

Particulate matter concentration prediction models have been researched, developed, and applied to data from various topography around the world. The characteristics of different topography make a model suitable for each specific topography. This dissertation proposes a novel method to predict particulate matter concentration with a diameter smaller than 10 microns in Nan Province of Thailand that integrates feature selection method, supervised learning model, and modified depth-first search algorithm. Unlike the traditional supervised learning models, the proposed method is able to accept multi-dimensional data as input which consist of particulate matter concentration, air pollutants, and air qualities. These features are the factors that influence particulate matter concentration prediction. The experimental results show that the proposed method performs better than other methods when predicting the concentration one hour ahead with no need of wind direction and wind speed data. The proposed method was developed with a general framework and could be applied to predict particulate matter concentration in Nan Province.