Testing spectral residual transformations on the manually labelled datasets to assess if the transformation could be useful in anomaly identification.
The notebook SR-explore.ipynb provides the exploration and the .csv files in this folder provide datasets that are standardized and then have a spectral residual algorithm applied.
Note that the idea came from Microsoft's use of spectral residual in their SR-CNN anomaly detection algorithm. The paper on SR-CNN can be found here.
The images\ folder contains various images referenced in the notebook file.
Overall, it could be useful to use the SR transformation with the LSTM as it looks like it could help highlight anomalies; however, as also discussed in the notebook, there are types of anomalous data that the transformation could make harder to detect. There was not sufficient time in this study to test the SR transformation but this is something that could be looked at in the future.