Grouping data in a multiline chart mean + outliers

Question

I have an existing multi-line graph that displays time series data about success percentages of nodes in a cluster in 5 minute intervals, there are more than 50 nodes in the cluster and the way this information is conveyed visually is very ineffective since it is quite impossible to tell the lines apart.

I figure, if I split the data into two groups, first one into which most of the nodes usually fall, would display normal operation and I could just draw its mean. And second one would contain just he outliers, the line(s) for nodes that are way below certain threshold or near zero, this would be much easier to interpret. Those two things are really what the people looking at those graphs care about anyway.

Can you recommend a technique I could apply to this? is this perhaps some common problem which I am failing to identify?

Technically it is just the matter of transforming my data, I understand, but since I am completely new to data analysis and much of statistics I wonder if there is some well known algorithm for doing this.

Answer 1

How you split the nodes into groups will depend on domain knowledge of what makes a node's series interesting. For comparison, there are general rule sets for signaling process control alerts, such the Western Electric rules.

However, I can suggest a technique for the data visualization. Plot the "normal" nodes as unlabeled, transparent gray lines, and plot the interesting nodes with one or more highlight colors. Labels and number of colors depend on the number of interesting lines, and sometimes the labeling is done through some kind of interaction (click or hover).

Some examples:

• Propublica Workers' Comp (hover labels)
• New York Times Arctic Ice (pictured above)
• Seattle Bike Blog (using mean as the highlighted line)
• Ann K. Emery Simplification and Emphasis