PREDICTING AIR QUALITY INDEX BASED ON METEOROLOGICAL DATA: A COMPARISON OF REGRESSION ANALYSIS, ARTIFICIAL NEURAL NETWORKS AND DECISION TREE

Abstract

Introduction: Air pollution can cause health problems on a global scale. Air quality predicting is an effective method to protect public health through early notification hazards of air pollution. The aim of this study is forecasting next day air quality index (AQI) in Tehran, Iran. Materials and methods: Various approaches such as multiple linear regression (MLR) analysis, decision trees (DT), and multi-layer perceptron artificial neural networks (ANN), feature selection with regression analysis before artificial neural networks (MLR-ANN) and feature selection with decision trees before artificial neural networks (DT-ANN) were used for forecasting next day AQI based on meteorological data and one and two days ago AQI. Root mean square error (RMSE) and correlation coefficient (CC) are used to assess models accuracy. Results: The results showed that forecasting of next day AQI by DT-ANN model is more accurate than others. Statistics indexes of this model such as RMSE and CC have been determined as 21.26 and 0.66 respectively. Using of DT for features selection because of reducing the number of inputs and decrease the model complexity was considered better than using the initial data. Conclusions: The applications of air quality forecasting methods could be applied for air quality management purposes and protect public health.

Keywords

• Air quality index
• artificial neural network
• decision Tree