Anomaly Detection in BACnet systems
Meeting 24
Title: K-modes optimization
Date: 28/03/2020
Time: 4 pm
Location: Zoom Conference Call
Attendees: Sofian Ghazali, Rahul Balamurugan, Zahid Kamil
Absences: None
Action Items: Decide on number of clusters
Meeting Agenda:
-
Is the elbow method good enough
-
Are there are any other methods that could help determine the number of clusters
Discussion and Notes:
The meeting was focused on determining if the clustering centroids are enough to represent the data. From the elbow method we noticed that 3 clusters is the optimal number but we know that three clusters would not suffice to represent the whole data. Therefore, we need other options. The silhouette score provides another means of determining the optimal number of clusters by measuring the dissimilarity between the clusters. Therefore, the more dissimilar the clusters the more they represent different aspects of the data. This is exaclty what we were looking for and we decided 14 clusters to be sufficient. Once decided we can now move onto the anomaly detection
Next meeting action items:​
-
Decide if there are any other changes to number of clusters
Meeting Adjournment: 6:00 pm
Meeting Minutes prepared by: Zahid Kamil