Probabilistic approaches for body fluid identification using RNA-data

Keywords
No Thumbnail Available
Issue Date
2018-12-03
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In forensics, body fluid identification is used to determine the type of body fluid found at a crime scene. The Department of Human Biological Traces of the Netherlands Forensic Institute uses an mRNA-based multi-marker approach. They use a categorical model to perform body fluid identification. The problem when using this method is the fact that no statement can be made about the uncertainty of the results. The use of a probabilistic model can solve this problem. Three probabilistic models from earlier research are reproduced and several other probabilistic methods are added to classify single body fluid samples. In actual casework, however, samples are made up of multiple body fluids. Alternative approaches based on multi-label classification, using the probabilistic methods created for the single body fluid classification, are implemented to classify these mixture samples. For the single body fluid sample classification the Multinomial Logistic Regression and Multilayer Perceptron models result in a classification accuracy of 88.2% and 91.2% and are therefore the models that classify the samples most accurately. The multi-label model using a Multilayer Perceptron in combination with the label power set method gives the most accurate outcomes on both the multiple body fluid samples (accuracy of 86.8%) as well as the single body fluid samples (accuracy of 89.3%). The multi-label model is implemented in a tool that can be used to analyze samples from actual casework.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen