Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 107 - Strengthening Forensic Science: The Contribution of Statistics
Type: Invited
Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
Sponsor: Advisory Committee on Forensic Science
Abstract #309197
Title: Data-Driven Decision-Making in Forensic Science Using Kernel- and Similarity Score-Based Methods
Author(s): Cedric Neumann* and Madeline Ausdemore
Companies: South Dakota State University and South Dakota State University
Keywords: kernel function; similarity score; biometry; high-dimension; inference; bayes factor
Abstract:

The past 4 decades have seen a push for more data-driven decision-making in forensic science. Inspired by the field of biometry, forensic scientists and statisticians have attempted to use kernel functions and other summary statistics based on the similarity between objects to design inferential models. Kernel functions and similarity scores are convenient as they can simplify the development of models for high-dimensional and complex evidence forms. However, the vast majority of the resulting models does not account for the correlation structures created by the pairwise comparisons of a set of objects. Hence, the use of these models can result in extremely misleading inferences. This talk will briefly review some of the issues related to the current generation of kernel- and score-based models, and will present some of the work that is underway at the South Dakota State University to propose logical and coherent inferential methods to forensic scientists.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program