Air Quality Index prediction using Deep learning for Lagos State in Nigeria

Abstract

The index for expressing air quality is known as the air quality index (AQI). It could be used to evaluate the effect of air pollution on a one’s health over a epoch of time which provides a guide to the community on the adverse health effects of air pollution around them.  This paper focused on developing a model for AQI prediction using Deep learning for Lagos state in Nigeria. The study acquired dataset from the OpenWeatherMap API which includes historical air quality and meteorological for Lagos State in Nigeria. Data was preprocessed by handling missing values, converting data types into numerical format using one-hot encoding. The study applied SMOTE technique to ensure balanced dataset. Four distinct Models such as LSTM, CNN, Prophet and SVR were utilized to determine the AQI of Lagos State. The results of balanced datasets used revealed LSTM provides the lowest MSE, RMSE and MAE values of 0.062, 0.249 and 0.149 respectively and higher R² value of 0.968 compared with the other model CNN, Prophet and SVR. The paper concluded that in the prediction of the AQI for Lagos State, LSTM outperformed other models such as CNN, Prophet Model and SVR on the validating metrics known as MSE, RMSE, MAE and R². The model obtained could be subjected to further research in other geographic regions, such as Delta State or other state by state level analysis which could be expanded to forecast other pollution indices at different levels.