Trace evidence is a broad category of valuable commonly encountered physical evidence. It is typically minute in quantity and may in fact be a complicated mixture of different materials. Mixtures of materials in trace evidence can be dust, gasoline, soil and other complicated combinations of matter. These classes of evidence can all be studied as patterns. Theoretically, no two patterns drawn from different sources will be exactly the same because all of the different processes that generated the patterns cannot be exactly duplicated.  Another form of physical evidence that can be analyzed as patterns is impression evidence.  Impression evidence includes such physical manifestations as   fingerprints, tool marks, footwear, tire tracks, and those impressions associated with firearms identification.

These types of forensic evidence are frequently challenged in court because the claims of uniqueness have been arrived at qualitatively rather then with statistical data. Thorough long term studies, utilizing controlled experimental conditions and standard statistical methods are not common.  This poses a problem in the wake of the Daubert decision in which the U.S Supreme Court rejected the Frye “general acceptance rule” governing the admissibility of scientific evidence. As a result, various forms of physical evidence have been the subject of Daubert based admissibility challenges.  By using statistical methods to analyze empirical experimental data one validates and lends objectivity to a field seen largely as subjective.

We are interested in addressing the need for a sound scientific basis for studying and comparing these categories of physical evidence. Our approach is to use methods of statistical pattern recognition. We are implementing progressively more sophisticated computer programs to carry out pattern recognition on trace evidence data we collect. As we collect data it is being stored in an ever expanding database to test and refine the pattern recognition algorithms we implement and to observe changes (if any) in the pattern data over time. Also, we are developing protocols for the use of multiple statistical methods to analyze the degree of similarity between arbitrary patterns in order to quantitatively define the term “matches”.

• See also:

Forensic Chemistry and Forensic Science Groups

• An excellent (though VERY technical) text that lays out the mathematically rigorous basis for the comparative sciences is:

The Nature of Statistical Learning Theory

By V. Vapnik

ISBN: 0387987800

° The full mathematical details of Professor Vapnik’s work are contained in:

Statistical Learning Theory

By V. Vapnik

ISBN: 0471030031