Neocortex 🧠

Search

Search IconIcon to open search

Conditional And Total Risk

Last updated Sep 25, 2022 Edit Source

The conditional risk of assigning an object $x$ to the class $\omega_i$ is ( given the missclassification costs): $$ l^i(x) = \sum_{j=1}^C \lambda_{j,i}p(\omega_j|x) $$ This means the average risk(expectation) over a region $\Omega_i$ is: $$ $$ $$ \begin{align*} r^i &= \int_{\Omega_i} l^i(x)p(x)dx \\&= \int_{\Omega_i} \sum_{j=1}^C \lambda_{j,i}p(\omega_j|x)p(x)dx \end{align*} $$

And the overall risk becomes:

$$ \begin{align*} r &= \sum_{i=1}^C r^i \\&= \sum_{i=1}^C\int_{\Omega_i} l^i(x)p(x)dx \\&= \sum_{i=1}^C\int_{\Omega_i} \sum_{j=1}^C \lambda_{j,i}p(\omega_j|x)p(x)dx \end{align*} $$

Classification The main purpose of classification algorithms is to choose the regions $\Omega_i$ so that each of those integrals is minimal.


Interactive Graph