Abstract:Abstract The Bayes’ theorem can be generalized to account for uncertainty on reported evidence. This has an impact on the value of the evidence, making the computation of the Bayes factor more demanding, as discussed by Taroni, Garbolino, and Bozza (2020). Probabilistic graphical models can however represent a suitable tool to assist the scientist in their evaluative task. A Bayesian network is proposed to deal with equivocal evidence and its use is illustrated through examples.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to evaluate the value of evidence in the presence of equivocal evidence. Specifically, the author explores how to use Jeffrey's Conditionalisation, a generalised form of Bayes' theorem, to deal with this uncertainty, and proposes the use of Bayesian Networks as an auxiliary tool to simplify the complex probability calculation process.
### Background and Motivation
In the fields of justice and forensic science, the evaluation of evidence credibility is very important. However, in many cases, scientists are not sufficiently certain about the truth of a proposition (for example, the characteristics of a scientific discovery), that is, there is "soft evidence" or equivocal evidence. In this case, the traditional Bayes' theorem is no longer applicable because the traditional Bayes' theorem assumes that all evidence is certain (i.e., "hard evidence"). Therefore, a method is needed to deal with this uncertainty in order to evaluate the value of evidence more accurately.
### Main Contributions
1. **Theoretical Extension**: The author derives a generalised Bayes factor formula that can handle equivocal evidence. This formula is based on Jeffrey's Conditionalisation and allows the probability of a hypothesis to be updated in the case of uncertain evidence.
2. **Application of Bayesian Networks**: In order to simplify complex probability calculations, the author proposes using Bayesian Networks to model and evaluate equivocal evidence. Bayesian Networks can not only intuitively represent the dependency relationships between variables, but also automatically update the probabilities of each node through probability propagation algorithms.
3. **Case Analysis**: Through specific case analysis, it is shown how to use Bayesian Networks to handle equivocal evidence and calculate Bayes factors, thereby evaluating the impact of evidence on hypotheses.
### Formulas and Methods
- **Bayes' Theorem**:
\[
\text{Pr}_0(H|E, I) = \frac{\text{Pr}_0(E|H, I) \cdot \text{Pr}_0(H|I)}{\text{Pr}_0(E|I)}
\]
where \(\text{Pr}_0\) represents the initial probability, \(H\) is the hypothesis, \(E\) is the evidence, and \(I\) is the background knowledge.
- **Jeffrey's Conditionalisation**:
\[
\text{Pr}_1(H, E) = \text{Pr}_0(H, E) \times \frac{\text{Pr}_1(E)}{\text{Pr}_0(E)}
\]
This formula allows the probability of a hypothesis to be updated in the case of uncertain evidence.
- **Bayes' Factor**:
\[
\text{BF} = \frac{\text{Pr}_1(R) \left[ \text{Pr}(E|H) \left( \text{Pr}(R|E) - \text{Pr}(R|\neg E) \right) + \text{Pr}(R|\neg E) \right] + \text{Pr}_1(\neg R) \left[ \text{Pr}(E|H) \left( \text{Pr}(\neg R|E) - \text{Pr}(\neg R|\neg E) \right) + \text{Pr}(R|\neg E) \right]}{\text{Pr}_1(R) \left[ \text{Pr}(E|\neg H) \left( \text{Pr}(R|E) - \text{Pr}(R|\neg E) \right) + \text{Pr}(R|\neg E) \right] + \text{Pr}_1(\neg R) \left[ \text{Pr}(E|\neg H) \left( \text{Pr}(\neg R|E) - \text{Pr}(\neg R|\neg E) \right) + \text{Pr}(R|\neg E) \right]}
\]
### Conclusion
Through the above