Definition

The information gain of a node $t$ of a decision tree is defined as $I G (t) = Q (t) - \frac{n _{L}}{n _{t}} Q (t_{L}) - \frac{n _{R}}{n _{t}} Q (t_{R})$ where $Q (t)$ is the impunity of node $t$ , $n_{t}$ is the number of samples used in $t$ , $t_{L}, t_{R}$ are children nodes of the $t$ .