Evaluation of Forensic Data Using Logistic Regression-Based Classification Methods and an R Shiny Implementation.

We demonstrate the use of classification methods that are well-suited for forensic toxicology applications. The methods are based on penalized logistic regression, can be employed when separation occurs in a two-class classification setting, and allow for the calculation of likelihood ratios. A case study of this framework is demonstrated on alcohol biomarker data for classifying chronic alcohol drinkers. The approach can be extended to applications in the fields of analytical and forensic chemistry, where it is a common feature to have a large number of biomarkers, and allows for flexibility in model assumptions such as multivariate normality. While some penalized regression methods have been introduced previously in forensic applications, our study is meant to encourage practitioners to use these powerful methods more widely. As such, based upon our proof-of-concept studies, we also introduce an R Shiny online tool with an intuitive interface able to perform several classification methods. We anticipate that this open-source and free-of-charge application will provide a powerful and dynamic tool to infer the LR value in case of classification tasks.