Information Statistic

Definition

Information Statistic (also Information Value) is a metric of the strength of a characteristic to separate a binary classification problem (such as good/bad classification in Credit Scoring).

Formula

$I = \sum_i ( p_b(S_i) - p_g(S_i) ) \log( \frac{ p_b(S_i) } { p_g(S_i) } )$

where

• $S_i$ is the value of the i-th class
• $p_b(S_i)$ is the probability (fraction) of bads for value class i
• $p_g(S_i)$ is the probability (fraction) of goods for value class i