An improvement proposed by Ross Quinlan that corrects for the disadvantage of Information Gain by taking into account the number of values a feature can take. Information Gain Ratio (IGR) is just the ratio of IG and Intrinsic Value (IV) which is defined as :

$IV(X, Y) = - \Sigma_{v \in values(Y)} \frac{|\{X_Y == v\}|}{|X|} \log_2 \frac{|\{X_Y == v\}|}{|X|}$ Thus, Information Gain Ratio is :

$IGR(X,Y) = \frac{IG(X,Y)}{IV(X,Y)}$