Recurrent Neural Networks for Analysis and Automated Air Pollution Forecasting

Time series prediction problems are a challenging type of predictive modelling case study as time series adds the involution of a temporal dependence among the input variables. Recurrent Neural Network (RNN)is a powerful type of neural network that able to handle sequence dependence cases. In this paper, we demonstrate analysis and automated air pollution forecasting using RNN. In our experiment, we built a distributed computing environment based on RHadoop, analyzed air pollution and presented visualization using HBase from historical data. Besides, we analyzed the short-term prediction of PM2.5 and measured the prediction accuracy based on mean absolute percentage error (MAPE) value. Finally, we utilized shiny to visualize the training result for optimizing parameters of RNN training module.