Multivariate Air Pollution Levels Forecasting

Abstract

The rising air pollution levels in a country are a matter of grave concern. For the development of measures to tackle air pollution, the forecasting of air pollutant levels becomes extremely important. Easier implementation of deep learning techniques in recent years has made the development of accurate forecasting techniques straightforward. In this paper, a multivariate forecasting framework is proposed to accurately predict various air pollutant levels in Indonesia. The pollutants include Particulate Matter 10 (PM 10), Carbon Monoxide (CO), Ground level Ozone (O3) and Nitric Dioxide (NO2). For each pollutant, a number of deep learning models have been separately trained and tested. The deep learning models include Multi Layer Perceptron (MLP), Convolutional Neural Networks (CNNs) and Long Short Term Memory (LSTM) networks. The model with the lowest errors on test data can be concluded as the most accurate on that pollutant and hence can be used for reliable future prediction.