The following tables and options are available for Anomaly visualizations.
Model Information table
Contains information about the data, including numbers of instances and features, and information about the Two-Step Cluster model on which anomaly detection is based, including the number of clusters or peer groups and their sizes.
Cluster Sizes chart
Provides a graphical representation of the relative sizes of the clusters or peer groups determined by the Two-Step Cluster model on which anomaly detection is based.
For each feature associated with at least one anomaly in a cluster or peer group, shows the number of anomalies associated with that feature within that peer group, as well as the average deviation index for those anomalies (on a 0-1 scale). This allows you to identify features contributing to anomaly status within each peer group. An important feature may contribute a large amount to one or more deviations or smaller amounts to many deviations.
The last line for each peer group or cluster shows the average residual deviation index for identified anomalies within that peer group. Large values for this residual, which is on a 0-1 proportion scale, indicate that anomalies within this peer group are not well explained by the identified features.
The Count value for the Residual line in each peer group shows the number of distinct anomalies identified within that peer group.
Peer Group Profiles Scale Features table
Shows means and standard deviations for each scale feature within each cluster or peer group. Comparing means for different peer groups helps you to see which features contribute to separation of peer groups. Within peer groups, instances far from a feature’s mean relative to its standard deviation would be more likely to be identified as anomalies.
Peer Group Profiles Categorical Features table
Shows counts and percentages of categories of each categorical feature within each cluster or peer group. Comparing distributions of category percentages for different peer groups helps you to see which features contribute to separation of peer groups. Within a peer group, instances that take on values other than the modal value would be more likely to be identified as anomalies.
Like your visualization? Why not deploy it? For more information, see Deploy a model.