I’d like to run anomaly detection on binary time series (port status activations 0/1) using NuPIC. Shall I change the representation of the data to scalar (like the amount of time the port is open a day and have daily data points), or it’s digestible in binary flavor, as well? Thanks!
For binary data, I would recommend using a category encoder with string categories of “0” and “1”. This will assure they have no overlap between them. I have an example of this in my cellular automata prediction project. Have a look in this video:
Also, I noticed when I built this project that I could have used better
w values for the category encoders I created. They should be adjusted so that
w takes up roughly half of the
n input space because there are only 2 possible inputs.
Fraud detection model with synthetic data set
Is code available? Can I implemented on my data?
Sensor_Id **Event_Date *Event_Hour:
Human_ presence : Represent the event In binary(0 indicate no human event) and (1 indicate a human event).
To detect anomalous behavior?
Thanks a lot