Analysis and prediction of air pollution using BiLSTM Conv1D

The source code for my master thesis.

The goal was to use Neural Networks to predict future values from historical data in specific locations.

In the context of this master thesis, 6 research papers were studied for the different types of neural network models on data related to air pollution and time series forecasting.

1. An LSTM-based aggregated model for air pollution forecasting
2. A Bayesian LSTM model to evaluate the effects of air pollution control regulations in Beijing, China
3. A Review of Neural Networks for Air Temperature Forecasting
4. Forecasting PM2.5 Concentration Using a Single-Dense Layer BiLSTM Method
5. A Hybrid Time Series Model based on Dilated Conv1D and LSTM with Applications to PM2.5 Forecasting
6. Deep-AIR: A Hybrid CNN-LSTM Framework for Air Quality Modeling in Metropolitan Cities

Furthermore, multiple different types of neural networks were developed in order to realize the final architecture of the proposed model. In addition, plentiful experimentation was conducted to determine the best hyperparameters for the model and was assessed both for its performance and accuracy but also for its generalization to more than one location around the globe.

---

Tools for data collection

1. OpenAQ
2. MeteoStat

---

Use of Libraries

1. Numpy
2. Pandas
3. Matplotlib
4. Seaborn
5. Scikit-learn
6. Datetime
7. Py-openaq
8. Meteostat
9. Tensorflow

---

Locations that were used for data collection and testing:

---

The proposed architecture of the BiLSTM-Conv1D network: