Using Gini Index to score a classifier tells you how close the classifier is to a perfect match vs how close it resembles random noise.

Gini Gain: the amount of Gini Impurity removed by performing a particular split.

Gini index

G = \sum^{C}_{i=1} p(i) (1-p(i))

C = total number of classes

p(i) = probability of picking the data point of class i

Gini Impurity

One subtracted from the gini index