Predicting Beijing Air Quality Using Bayesian Optimized CNN-RNN Hybrid Model

Abstract

Poor air quality impacts lives around the world every day, causing problems that range from respiratory infections to mental illnesses to death. Being able to reliably predict when air quality will be the worst will allow organisations to take action and precautions in order to reduce incoming pollution or to keep people safe. In this paper, we introduce a Bayesian Optimized CNN-RNN hybrid to tackle this problem. We chose this solution in order to avoid the problems that arise from manual hyperparameter adjustment commonly found in neural networks. Training and applying this model to the Beijing Multi-Site Air Quality Dataset, we compared it to other traditional machine learning algorithms such as ARIMA, CNN, and RNN. In the end, the BO-CNN-RNN was able to outperform the other models, even better as predictions went further into the future.