Author(s)

Piotr A. Kowalski, Kasper Sapała & Wiktor Warchałowski

Abstract

The World Health Organization (WHO) estimates that air pollution kills around 6.5 million people around the world every year. The European Environment Agency, in turn, points out that about 50,000 people die annually in Poland due to this. PM10 pollution arises in the form of smog (smoke and fog) and is an unnatural phenomenon created by adverse weather conditions and human activity. The aim of this article is to assess the possibilities of tasking modern neural networks to predict PM10 air pollution levels in the following hours of the subsequent day. In evaluating the prediction task, several types of error are considered, and machine learning algorithms and structures are utilized as learning models. Of note, the algorithm selected for stochastic optimization is a form of convolutional neural networking and deep learning neural networking that is used in machine learning when considering Big Data issues. The obtained results were then analysed and compared with other methods of prediction. As a result of this research, the proposed convergent neural network could be used effectively as a tool for calculating detailed air quality forecasts for the subsequent 24-h period.

Keywords

air pollution prediction (forecasting), big data, convolutional neural networks, machine learning, regression task, neural network, particulate matters.