Air Pollution Matter Prediction Using Recurrent Neural Networks with Sequential Data

Abstract

Air pollutants such as fine dust and ozone are important factors in human health management. In this work, the future air quality of Daegu metropolitan city is predicted by using the past air quality data. Due to the time series nature of the data, we use recurrent neural networks for the experiments. The data is measured in units of one hour using various air quality sensors. Experiments were performed based on length of input data (time step) in order to obtain the optimal length. Various optimization functions and neural network structure were also investigated. The prediction accuracy of fine dust was found to be the most predictable among other environmental pollutants. Also, it was observed that learning models for nearby areas can be used to predict similar pollutant in another area without having to go through a separate learning process.