Leveraging Long Short Term Memory Neural Networks in Air Pollution Prediction

Abstract

Air pollution is a serious environmental health hazard which contribute 6.7 Million premature deaths worldwide annually. Studies have shown that increased exposure to air pollution exacerbate cardiovascular and respiratory diseases such as heart failure, stroke, chronic obstructive pulmonary disease, lung cancer and pneumonia. Major factors that determine air pollution are PM2.5, carbon monoxide(CO), Sulphur dioxide(SO2) and Ozone(O3). To effectively control air pollution,monitoring and forecasting is vital. Our goal is to predict the level air pollution using Long Short Term(LSTM) Deep Neural network algorithm. This method has emerged as a superior technique in predicting air quality index(AQI) compared to other machine learning algorithms. We apply the LSTM model to predict the concentration of PM2.5 in the city of Nairobi which is a major contributor to health problems. We also quantify the effects of air pollution with the number of new admissions of patients to hospitals with respiratory diseases. Sensor data collected at GeoHealth Hub was split into training, validation and testing and train the model with Adam optimization algorithm. The Root Mean Squared Error (RMSE) was used to evaluate the model performance. We compared the performance of the proposed model with the baseline Random forest model and the results showed that LSTM model performed performed better than Random forest Therefore, leveraging LSTM neural network algorithms for fitting and predicting complex and flexible models are increasing, however not widely used in air pollution. The study utilized this technique to obtain a complete understanding of the levels of air pollution and also generate hypotheses for future studies that would inform the development of strategies to mitigate the risk associated with air pollution.