Artificial Neural Network for the Prediction of Particulate Matter (PM2.5)

In this paper, artificial neural network model developed for the prediction of PM2.5 particulate matter is discussed. Along with artificial neural network, multiple linear regression model is developed for the prediction. Both the techniques are compared for the prediction of PM2.5. The data for the training and testing is collected from OpenAq platform. The result shows that observed and predicted values are in close agreement. Artificial neural network is proved to be better than multiple regression model for the proposed application. Proposed method can also be used for data imputation technique for pollutant dataset.

I. Introduction

Increase in industrialization, smart cities, increase in traffic and larger energy consumption had resulted in tremendous air pollution. The air consists of hazardous pollutants like Sulphur dioxide SO₂, Nitrogen dioxide NO₂, Carbon monoxide CO, Ozone O₃, and Particulate Matters PM_2:5 and PM₁₀ which are originated from various sources. These pollutants have many health hazards, hence air pollution is recognized as a crucial factor. Hence, by predicting air pollution in advance can have certain advantages which includes preparedness & policy formation for management and control in fields which directly or indirectly depends on air pollution. Therefore, there is increasing demand for air quality forecasting. Traditional approaches involve various statistical approaches that demand a great amount of computer resources. Recently due to advancement in computational power various novel approaches for prediction of air quality have been initiated. Artificial Neural Network (ANN) is one of such approaches that produce state of art results for classification and prediction. Recently the neural networks are used for power quality disturbance detection and classification [1], neural network is also used for various applications of medical as well as different engineering fields [2].