Using common sense, you'd review the suspicious transactions with higher dollar amounts first.
"Rob" <email@example.com> wrote in message news:firstname.lastname@example.org... >I have an optimization problem that I would appreciate some help on. > > First for the background... > > 1. I have millions of purchase transactions. > 2. Each transaction is scored for its probability of being fraudulent. > To keep things simple, we will say that a transaction that has a 63% > chance of being fraudulent is scored 63, 22% is scored 22, etc., with > the score ranging from 0 to 100. > 3. Each transaction also has a dollar amount associated with it. As > you might expect, this ranges from under one dollar to over a million > dollars. > 4. The distribution of the transactions by score is almost logarithmic > in that for every high-risk transaction there are literally thousands > of low risk transactions. In other words, the vast majority of > transactions are not fraudulent. > > Now the objective... > > The objective is to: > 1. Minimize the dollar amount of losses due to fraud. > 2. While minimizing the number of transactions that must be reviewed. > > I have Base SAS 9.1.3 and SAS Stat available. Your thoughts are deeply > appreciated. >