Likelihood Ratio

Use-case

The Likelihood Ratio defines a Discriminant Function without biases (priors) in the context of a Bayesian classifier.

Mathematical Definition

For a two class case,

$$\boxed{LR(\mathbf{x})=\frac{p(\mathbf{x}|s=1)}{p(\mathbf{x}|s=2)}>\theta \Rightarrow \text{Decide for }s=1,\text{ else for }s=2.}$$

Variants of the threshold:

• Maximum Likelihood Decision

$$\theta =1$$

Simple decision

• Bayes Decision

$$\theta =\frac{P(s=2)}{P(s=1)}$$

Makes use of Prior

• Bayes Risk Decision

$$\theta =\frac{{\lambda }_{12}-{\lambda }_{22}}{{\lambda }_{21}-{\lambda }_{11}}\cdot \frac{P(s=2)}{P(s=1)}$$

Use Prior and Cost

Note: ${\lambda }_{12}$ is the cost of deciding state $1$ given state $2$ is true, so an and so forth.

Note: The threshold has same units as the $y-$axis.

---

References

Source