Anomaly Detection in BACnet systems
Meeting 27
Title: Testing k-modes
Date: 01/04/2020
Time: 6 pm
Location: Zoom Conference Call
Attendees: Sofian Ghazali, Rahul Balamurugan, Zahid Kamil
Absences: None
Action Items: Test K-modes
Meeting Agenda:
-
Decide on distance metric
-
Determine distance accuracy
Discussion and Notes:
The meeting was focused on which distance metric to use and how would we fit all of this into the GUI. The distance metric based on recommendations was the Jaccard distance. The jaccard distance works well with categorical data as shown and compared to the other distance metrics the threshold value chosen would be the best since it is right in the middle (0.45-0.5). With the testing done and obtaining a 98% accuracy we knew that the model was fed very little data and that it is obviously overfitting. Therefore, for future uses more data is the better option and this is why data is the new business. Next meeting we should see what we get with other clustering numbers
Next meeting action items:​
-
Move on to K-means
Meeting Adjournment: 6:30 pm
Meeting Minutes prepared by: Zahid Kamil