Re: Statistical Analysis on the IBM i

On 03-Dec-2013 06:42 -0800, Charles Wilt wrote:

On 02-Dec-2013 12:12 -0800, Charles Wilt wrote:

<<SNIP>> In particular, I'm interested in running some analysis in
near real time at the order line item level; possibly with models
built offline
<<SNIP>>

<<SNIP>> Basically, I have a requirement: Identify "suspicious"
orders which include orders of unusual size, orders deviating
substantially from a normal pattern, and orders of unusual
frequency.
<<SNIP>>

I am not sure if the following scheme might be useful to achieve the requirements in the described scenario?:

Create an actual or effective Materialized Query Table (MQT) that is a summary of the historical data using desired aggregates. Use that somewhat static data as reference-data for effecting the sanity-check of the attributes of the new data\order. Schedule a refresh of the summary-data after some period of time or number of additions [or changes] to the [apparently historical] detail-data, or after some deviation that is deemed acceptable to be applied to the detail-data.