Excessive levels of air pollution are one of the biggest threats to health of humans and natural environment. Thus, it is particularly important to accurately forecast air pollution in order to minimize its hazardous effects and support multiple day-to-day decision making in various systems. As part of this work, the existing state-of-the-art for predicting the level of air pollution were reviewed. Subsequently, theoretical foundations of Spiking Neural Networks (SNNs), which were used to conduct the study, are also presented. The author of the thesis proposed and implemented three architectures of SNNs based on recently introduced snnTorch package. In order to test the prepared implementations, the real dataset of Particulate Matter 2.5 (PM2.5) air pollution for Warsaw was selected and preprocessed. As the results of the experiments showed, the selected spiking neural networks performed better than the non-spiking artificial multilayer perceptron.
├── README.md
├── data
│ ├── processed <- Final, processed data in csv files.
│ └── raw <- Raw data in text files.
│
├── notebooks <- Jupyter notebooks.
│ ├── data_processing.ipynb
│ ├── dataset_properties.ipynb
│ ├── profiling.ipynb
│ └── training_stats.ipynb
│
├── src <- Source code for use in this project.
│ ├── __init__.py
│ ├── error_measures.py
│ ├── graphs.py
│ ├── models.py
│ └── train_model.py
│
└── requirements.txt