Decision Analysis in Forensic Science

Abstract

Forensic scientists are routinely faced with the problems of making decisions under circumstances of uncertainty (i.e., to perform or not perform a test). A decision making model in forensic science is proposed, illustrated with an example from the field of forensic genetics. The approach incorporates available evidence and associated uncertainties with the assessment of utilities (or desirability of the consequences). The paper examines a general example for which identification will be made of the decision maker, the possible actions, the uncertain states of nature, the possible source of evidence and the kind of utility assessments required. It is argued that a formal approach can help to clarify the decision process and give a coherent means of combining elements to reach a decision.

Likelihood Ratio Calibration in a Transparent and Testable Forensic Speaker Recognition Framework

“…Therefore, an optimization process focused on the rate of misleading evidence under H d will lead to a calibration loss. While the reduction of the rate of misleading evidence and the limitation in such numerical LR values are important points in forensic systems, an optimal process for accomplishing this objective would imply the calibration of the system LR values and then the use of costs or utilities in the decision process by the fact finder, as has been recently proposed in forensic science [26].…”

“…First, it allows the forensic scientists to estimate and report a meaningful value to the court [16]. Second, the role of the scientist is clearly defined, leaving to the court the task of using prior judgements or costs in the decision process [26]. Third, probabilities can be interpreted as degrees of belief [27], allowing the incorporation of subjective opinions as probabilities in the inference process in a clear and scientific way.…”

“…Recent approaches for speaker recognition evaluation have proposed the use of application-independent metrics such as C llr [21], where application, as defined in [21], is the set of prior probabilities and decision costs involved in the inferential process [26]. C llr is a single scalar value defined as: .…”

“…C llr presents several interesting properties. First, the LR values are evaluated in an application-independent way, which in forensics means case-independent, where different costs and priors may be involved in the decision process of each different case [26]. Second, as a single scalar value, C llr is very useful in order to easyly compare and rank systems.…”

Suspect-Adapted MAP Estimation of Within-Source Distributions in Generative Likelihood Ratio Estimation

“…Therefore, this value allows not only to discriminate between suspects, but also to infer posterior probabilities, or confidences [5], in order to take decisions in a transparent and scientific way. Second, the role of the scientist is clearly defined, leaving to the court the task of using prior judgements or costs in the decision process [7]. Third, probabilities can be interpreted as degrees of belief [8], allowing the incorporation of subjective opinions in the inference process.…”