Gini Index

Measures statistical dispersion.

0 → perfect equality, all values same class 1 → maximum ineuqality

Based on Lorenz Curve.

$\mathrm{Gini}(D)=1-{\sum }_{j=1}^n{p}_j^2$

$\Delta {\mathrm{Gini}}_A(D)=\mathrm{Gini}(D)-{\mathrm{Gini}}_A(D)$

Attribute with minimum Ginig Index is used.

Example

Disadvantages

• Biased to multi-valued attributes
• Has difficulties when number of classes is large
• Tends to favor equal-sized partitions and purity in both of them