Consider a dataset with the following classes in the parent node: $Y=[A,A,B,B,B]$ Let's assume that a split divides this data into two child nodes:

Left child node: $Y_l=[A,A,B]$

Right child node: $Y_r=[B,B]$