[rules-users] Some Fusion clarification

[Peter Williams]

I interested in general commentary on this topic.

How strong is the community's work when augmenting the realtime rule scanning (over potential fraud events) with statistical support package? E.g. something applying baysian/statistical logics?

The rules engine is not that interesting for fraud management unless augmented with such a package - one which can continually rebalance the percentiles that correspond to different risk bands - for some cumulative risk score over the various fraud predictors for the compliance/remediation program of some industry area. Typically, such predictors are velocity counters (e.g. how many claims of type X are submitted per office, or per individual), and the percentiles are rebalanced nightly to help auto-learn over time what the norms are for each velocity metric.
 
-----Original Message-----
From: rules-users-bounces at lists.jboss.org [mailto:rules-users-bounces at lists.jboss.org] On Behalf Of djb
Sent: Wednesday, June 23, 2010 4:03 AM
To: rules-users at lists.jboss.org
Subject: [rules-users] Some Fusion clarification


Hi all,

I'm trying my hand at CEP for fraud detection (analysing incoming claims to determine whether they are suspicious), and have 2 question/clarifications.

1.
Is the only difference between a fact and an event that an event has a timestamp/duration?

I have medical claims coming in, and basically, that is my only event type; say, ClaimReceivedEvent.

Now, I've got my rule engine working, where Claims are facts.  Will I now just insert my facts as events, and continue as normal, using accumulate and sliding window syntax?  

2.
Most of my rules will have to do with comparing complex variables from the incoming claims to historical averages.  (I assume this is the main use of Drools Fusion, in general)

Should I just compute these averages in java and insert them into the WM?  

If I want to compare monthly averages, (lets say, this month a provider claims 50% more than last month), am I meant to calculate this beforehand? 
Or, do I use pseudo-time to pump through historical events, all the while firing rules to update the average, until I have processed all of the historical data into these running average variables, and then switch to real-time?

I'm having trouble with this big picture.


Thanks,
Daniel











Thanks,
Daniel



--
View this message in context: http://drools-java-rules-engine.46999.n3.nabble.com/Some-Fusion-clarification-tp916521p916521.html
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users