How to solve crime with science: Evidence evaluation with statistics and machine learning

Peter Vergeer

Netherlands Forensic Institute, R&D data team of the Chemical and Physical Traces division

Resumen/Abstract

This lecture is an introduction into the general concepts involved in evaluation of forensic evidence by statistics and machine learning. It starts from first principles, introducing the likelihood ratio (LR) as a measure for evidential strength, and then introduces the two-level model (TLM) of evidence evaluation. This is the most-used model for forensic inference on continuous data. As a practical application, such a TLM LR-system trained on glass data will be introduced, with a focus on measuring performance and the need for post-hoc calibrating. After this, the talk will end with a brief look into the merits that LR-systems trained as neural nets could bring

Curriculum ponente

Peter Vergeer is a research scientist in the fields of forensic statistics and data science at the Netherlands Forensic Institute. He received his M.Sc. in research methods and statistics for Psychology from Leiden University in 2010 and his Ph.D. in Chemistry from Utrecht University in 2005. His research is mainly concentrated on computer-based methods for evaluation of strength of evidence for forensically relevant data. Amongst others, he is author of a number of research papers, ranging from theoretical frameworks for how we calculate and measure evidential strength, to applying state-of-the art methods to calculate evidential strength for a number of evidence modalities.