Beijing PM2.5 Prediction Algorithm Based On Machine Learning

As the level of fine particulate matter(PM2.5)and other air pollutants continues to rise,the issue of air pollution continues to be a big problem in China and especially in Northeast China where the pollution often gets severe.This thesis uses Machine Learning methods to predict the concentration levels of PM2.5 in Beijing using PM2.5 concentration data and meteorological data collected in the year 2015.This thesis used five machine learning classification algorithms available in Machine Learning;namely:Naive Bayes,Multinomial Logistic Regression,Sequential Minimal Optimization,k-Nearest Neighbor(k-NN)and Random Subspace.PM2.5 predictive models are built using the WEKA implementation of these classification algorithms and the data used therein was pre-processed and visualized using R Studio.Through the research performed we see that Random Subspace using Logistic Model Tree(LMT)as the base tree classification algorithm had the best performance with prediction accuracy of 70.92%and Humidity among other meteorological factors had the biggest effect on the level of PM2.5 in the atmosphere.This study aims to help with the prediction of PM2.5 so the general population can be informed and take appropriate measures to protect their health.