This code provides the implementation details of a Feature Drift Explanation (FDE) module responsible for identifying drifts in a Deep Learning (DL) regression problem. The use case relates to a cyber-security attack that targets the readings of specific sensors supplying data for a 1-dimensional Convolutional Neural Network (1D-CNN) that predicts CO2 concentrations to aid Heating, Ventilation, and Air Conditioning (HVAC) systems in their activations. As such, finding the compromised sensors via FDE allows for a targeted mitigation strategy that potentially reduces the HVAC system's energy consumption and maintains the occupant's comfort.

The code documentation aligns with the content of the manuscript, which is accepted as part of the International Wireless Communications and Mobile Computing Conference 2024 conference. The pre-print for this manuscript can be found using the following link: https://arxiv.org/abs/2403.13023

Before running the code available in the FDE_Notebook.ipynb, it is imperative to retrieve the data used for this paper. This can be achieved using the following steps:

1. Download the data used for this work, which can be found using this link: https://zenodo.org/record/3774723#.ZGEhAHbMKUl. In this repository, the authors provided the prepare.py file to generate continuous chunks of data (60 minutes). This produces two files: 1) continuous_sections_60_train.pickle and 2) continuous_sections_60_test.pickle.
2. Create the data folder in the root directory, which is on the same level as FDE_Notebook.ipynb. Move the two generated files, continuous_sections_60_train.pickle and continuous_sections_60_test.pickle to this directory.

This concludes the steps required before running the FDE_Notebook.ipyb notebook. This notebook includes extensive explanation about the steps involved in the FDE process.

We also included a models directory. In that directory, two models, which were used to generate the paper's results, are available:

• cnn_model.pt: the 1D-CNN model used for CO2 level predictions.
• ae_full_model.pt: the auto-encoder used for retrieving the activations' latent representation.

Please feel free to contact me for any questions or research opportunities.

• Email: [email protected]
• Gihub: https://github.com/ibrahimshaer and https://github.com/Western-OC2-Lab
• LinkedIn: Ibrahim Shaer
• Google Scholar: Ibrahim Shaer and OC2 Lab