Interesting article from IEEE Spectrum:
During two years of deliberation by the National Academy’s forensic science committee (of which I was a member), a troubling picture emerged. A large part of current forensics practice is skill and art rather than science, and the influences present in a typical law-enforcement setting are not conducive to doing the best science. Also, many of the methods have never been scientifically validated. And the wide variation in forensic data often makes interpretation exceedingly difficult.
So how might greater automation of classical forensics techniques help? New algorithms and software could improve things in a number of ways. One important area is to quantify the chance that the evidence is unique by applying various probability models.
Computational forensics can also be used to narrow down the range of possible matches against a database of cataloged patterns. To do that, you need a way to quantify the similarity between the query and each entry in the database. These similarity values are then used to rank the database entries and retrieve the closest ones for further comparison. Of course, the process becomes more complicated when the database contains millions or even hundreds of millions of entries. But then, computers are much better suited than people to such tedious and repetitive search tasks.