auto classification

[NYSLPE]

Hi,
I would like to get your opinions on auto classification. Our agency is considering using strictly auto-classification on tenprint transactions. Since our latents cases will be searched against a database of nothing but auto-classified patterns, we’re skeptical. Our experience with the FBI’s auto classification shows that it over- references…a lot. We seem to have to wade through a lot of false candidates….Anybody else notice this?

Becky

[Ann Horsman]

If you are refering to 'lights out' ten print QC then I am against it personally. There is debate about it here in VT with our shared AFIS database.

[ER]

In Arizona, we went to complete auto classification a couple of years ago. In addition to new prints being auto-classed, all of the classifications on archived prints were deleted and replaced with new, auto-generated classifications. And yes, it over-references a lot. However, I have yet to see a completely incorrect classification. They may call a left-slant loop a whorl, but they also reference it to a left-slant loop. I've yet to see them get it completely wrong.

I understand why there is a push for lights-out classifications. However, I think that a happy medium between completely lights-out and completely manual would be for manual review of any classification that gets a reference pattern or an unknown pattern from the computer. This process would be mostly automatic but also keep our databases cleaned up and help with getting relevant search results.

[NYSLPE]

ER,

It sounds like a very reasonable idea to have human intervention when the computer can’t decide. Is that the route your agency is thinking about going?

[NYSLPE]

We are concerned that with lights out auto-class, the database will get so watered down that we will have a more difficult time getting hits on our latent cases. Nowadays, when we search against the FBI we have to narrow our search way down and search multiple times to get under the penetration rate. Is that going to be our future when our system goes to autoclass? (I don’t mean to bash the FBI’s system, but my experience with autoclass is mostly with the FBI)

Also, will we be seeing so many false candidates on our reverse/cascade searches we won’t have time for other cases?

Becky

[ER]

It sounds like a very reasonable idea to have human intervention when the computer can’t decide. Is that the route your agency is thinking about going?

NYSLPE,

No. That was just my hope of what might happen. I doubt that human intervention will ever make it back into the process. That costs money, and nobody has any.