Deep learning for prediction of air pollution trend for four monitoring stations in Delhi, India.
The experiments feature different deep learning models such as FNN and LSTM based models. The experiments also feature multivariate and univariate analysis using different training strategies with Bi-LSTM model (best) for multistep ahead prediction of PM2.5 values. The experiments further provide one month ahead forecast of PM2.5 values with uncertainty quantification.
We have a unified code for all four stations with proper comments.
- The python notebook for implementation can be found here: Implement
- The python notebook for data visualization can be found here: Visualize
The updated data used in experiments can be found here: DATA
Sample results (30 runs) for different stations using different models and training strategies can be found here: results. The analysis and plots for different monitoring stations can be found here: plots