I am currently doing research on anomaly detection and currently I am facing the following problem:
My input data has only one label (anomalies) and the associated data has a very high dimension and is binary (true/false). At the moment it seems to me that I could achieve the best result with the One-class SVM. One restriction I have is that the algorithm should be executable on a µC or a light hardware accelerator.
Do you have any suggestions for algorithms or papers I should look at?
Thanks a lot!
Edit: I also had a look at PCA but it seems to me that the compute is quiet heavy.