Anomaly detection for unlabeled data

Help
#1

Hi, all!
I am trying to train the unlabeled data using anomaly detection method for some prediction or detection purpose. The reason of using Edge Impulse but not the Python environment like Jupyter or Colab is that I am using Arduino Nano and several sensors to collect the data. Once finish the data collection, I am planning to do the anomaly detection for training and compile the model to the Arduino board for further detection.
The problem which I currently facing is I can’t find the way to train the model without label. So may I know is it possible to use anomaly detection with unlabeled data in Edge Impulse?
Any solution or guidance would be most appreciated.
Thank you.

Sincerely,
Yin

#2

Hi @KeYin,

For Edge Impulse, your data must have labels. If you are just doing anomaly detection (and not classification), you can give all your samples a single, arbitrary label (such as “asdf” - the label string will not matter).

1 Like
#3

Hi @shawn_edgeimpulse ,
I have tried your suggestion method. It works for me now.
Thank you very much for your help!

Thank you.

Sincerely,
Yin

1 Like