Comparison of Different Machine Learning Methods to Forecast Air Quality Index

Abstract

In recent years, more and more people have been plagued by respiratory diseases. The air quality, which is characterized by inhalable particles and fine particles, has attracted increasing attention. Accurately monitor and forecast the quality of air could not only help the government conduct interventions to alleviate the air pollution earlier, but also alert relevant people who suffer from respiratory diseases. In order to develop effective Air Quality Index (AQI) prediction models, this paper compared the performance of different Machine Learning (ML) methods and feature selection methods. First the air quality data in Beijing from 2016 to 2017 were collected. Then Multi-Linear Regression (MLR), Random forest Regression (RFR), BP Neural Network (BPNN) and Support Vector Regression (SVR) algorithm were trained on 10-fold cross validation. Correlation coefficient (R), mean absolute error (MAE) and root mean square error (RMSE) were used as evaluation metrics. The experimental results showed that the performance of SVR and BPNN were similarly well. MLR had the worst performance, which was possibly caused by a small feature dimension, and RFR had higher accuracy and better generalization capability than the other models, probably because the algorithm of regression tree in random forest included the interaction of variables.