Air pollution in urban environments has risen steadily in the last several decades. Such cities as Beijing and Delhi have experienced rises to dangerous levels for citizens. As a growing and urgent public health concern, cities and environmental agencies have been exploring methods to forecast future air pollution, hoping to enact policies and provide incentives and services to benefit their citizenry. With greater computing power in the twenty-first century, using machine learning methods for forecasting air pollution has become more popular. This project investigates the use of the LSTM recurrent neural network (RNN) as a framework for forecasting in the future, based on time series data of pollution and meteorological information in Beijing. Our results show that the LSTM framework produces equivalent accuracy when predicting future time stamps compared to the baseline support vector regression for a single time stamp. Using our LSTM framework, we can now extend the prediction from a single time stamp out to 5 to 10 hours in the future. This is promising in the quest for forecasting urban air quality and leveraging that insight to enact beneficial policy.
"We see that we are able to forecast pollution within the same error margin as prediction for a single future time-step."
"LSTMs are a good model given the scope of the data we obtained, since they preserve the errors in a gated cell"
"Given this paper's results that we can predict air pollution accurately up to ten hours in advance, this can be extremely useful for city policy, and more specifically, for dynamic public transit pricing."
"In addition, the results of this paper can help in the management of deployed sensors, primarily in decreasing power consumption. We can now help cities save money by keeping their sensors on only in six to ten hour increments, rather than 24/7..."