The probability of determining the probability of matching fingerprints

I’m on the periphery of the forensic science/law enforcement world.

By CBS Television – eBay itemphoto frontphoto back, Public Domain,

Yes, I have completed training on forensic face recognition, but that doesn’t qualify me as an expert in courtroom testimony. (Forensic face recognition expert testimony isn’t admissible in court anyway, but you get the idea.)

But even I am well aware that the forensic world changed dramatically in 2009.

Before 2009, the dialog below only represents a slight exaggeration.

Question: Why do you say that these two fingerprints belong to the same person?

Answer: Because I said so.

After 2009, specifically after the release of what is called “the NAS report,” there has been an effort to make forensic science…a science.

Ideally, this means that when a fingerprint expert testifies in court, the expert can state that there is a 99.9978% probability that two fingerprints belong to the same person. Or something like that.


We’re not there yet, as this 2017 IAI position paper implicitly states.

It is the position of the IAl that examiners are encouraged to articulate conclusion decisions as specifically as possible, as to not overstate decisions regarding source attribution. In addition to stating conclusions, examiners are encouraged to state the basis for resulting conclusions; including the associative strength and limitations. The strength and limitations of conclusions may include the quality and quantity of data, the validity of method/mathematical model used, and the repeatability of the conclusion. Examiners are encouraged to continually reassess methods and/or mathematical models used to arrive at the best conclusions possible.

International Association for Identification, Position Statement on Conclusions, Qualified Opinions, and Probability Modeling, February 5, 2017.

In other words, while the IAI discourages the use of the old “Because I said so” articulation, the conclusions stated in court lean more toward qualitative rather than quantitative criteria. There’s not a probabilistic model for fingerprints.

Or, as Mike French notes, there’s not a publicly available probabilistic model.

As organizations like the Center for Statistics and Applications in Forensic Evidence (CSAFE) explore the viability of statistical modeling in pattern evidence disciplines, they will probably notice that AFIS (automated fingerprint identification system) vendors have already done decades of research, and those vendors have fielded operational systems, to solve the same type of problem forensic researchers are now investigating. 


French notes a number of challenges to using AFIS vendor data to derive probabilistic models for fingerprints, but the chief challenge is the fact that the AFIS vendor data is proprietary and therefore carefully guarded. After all, AFIS vendors understandably don’t want their competitors to be able to reverse engineer their algorithms.

If you read French’s article, you’ll see that even if the AFIS vendors made all of the relevant data available, significant testing would still have to take place before reliable, fit for purpose probabilistic models can be created.

Are there other ways to develop a probabilistic fingerprint model? Maybe, but these would require (among other things) access to a lot of fingerprints, and considering the resistance of privacy advocates to biometric collection—even when such collection can mitigate privacy advocate concern about biometric inaccuracy—the chances of collecting a bunch of fingerprints for a probability study are approximately 23 (and me?) in 7 billion.


Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s