High AUC and model accuracy but low gains in top decile

[kanika harjai]

Hello

I have a case where I am building logistic regression model for segments of a population. I have one model for the overall population, and one model each for two mutually exclusive sub-segments. My overall model has good accuracyand gains chart shows that 60% of True positives have been captured in the Top Decile. However, when I compare it to the segmented models, one of the segments has good accuracy and AUC values, but the top decile captures only 30% of true positives according to the gains chart. That appears to be significant drop in the top decile gains for that segment, and I am interested in maximizing that.

Can you please help me understand under what scenarios does this happen Why is the accuracy,AUC good but the %TP captured in top decile lower

Thanks!