Pollutants in the air lead to degradation in the air quality which leads to global warming and unpredictable heatwave occurrence. Air pollution has been a leading cause of respiratory diseases and premature deaths. To forecast AQI and temperature, a multivariate approach based on AR-Net and Temporal Fusion Transformer (TFT) is proposed. A combination of atmospheric and meteorological variables as input features to train the model, including temperature, humidity, wind speed, and pollutant concentrations (PM2.5, PM10, SO2, NO2). Open-source dataset of Air quality in India, is used to train the models and evaluate the experiments. With an MAPE of 7% for AQI and MAPE of 4%for heatwave prediction, it is concluded that the results demonstrate appreciable accuracy in predicting AQI and the occurrence of heatwaves.
-
Install the dependencies- pip install requirements.txt
-
Data- Data is provided in the github.
-
Run the jupyter notebooks.
- PyTorch
- NeuralProphet
- PyTorch Lightnining
- TensorFlow
- Pandas
- NumPy
The significant contribution of the work can be noted as the following.
• Identification of future air quality and temperature prediction with ease.
• Comparison of various deep learning techniques to select the best model for each use case, i.e., AQI and heatwave prediction.
- Abhinandan Roul - @abhinandanroul
- Shubhaprasad Padhy - @blixard
- Sambit Kumar Sahoo - @samSambit12
- Ayush Pattanayak - @ayush06032002