I ran across this blog post by Tim Bass on Complex Event Processing dubbed "Orwellian Event Processing" and it struck a nerve.
The rules we build into products like Netcool Omnibus, HP Openview, and others are all based on simple if-then-else logic. Yet, through fear that somebody may do something bad with an event loop, recursive processing is shunned.
In his blog, he describes his use of Bayesian Belief Networks to learn versus the static if-then-else logic.
Because BBNs learn patterns through evidence and though cause and effect, the application of BBNs in common event classification and correlation systems makes total sense. And the better you get at classification, the better you get at dealing with uncertainty.
In its simplest form, BBNs output a ratio of occurrences ranging from 1 to -1 where 1 is 100 percent that an element in a pattern occurs and -1 or 100% that an element in a pattern never occurs.
The interesting part is that statistically, a BBN will recognize patterns that a human cannot. We naturally filter out things that are obfuscated or don't appear relevant.
What if I could have a BBN build my rules files for Netcool based upon statistical analysis of the raw event data? What would it look like as compared to current rule sets? Could I setup and establish patterns that lead up to an event horizon? Could I also understand the cause an effect of an event? What would that do to the event presentation?
Does this not open up the thought that events are presented in patterns? How could I use that to drive up the accuracy of event presentation?