Each line represents a different number of attributes (the number of samples is fixed to 200) and each graph is associated to a different percentage of one values.