High AUC, moderate precision and recall in ~1:1 ratio: the best model that balanced precision and recall?

Question

I have trained a feedforward neural network to classify imbalanced dataset (pos:neg=67:1) with the following performance:

High AUC : 0.964

Moderate Precision : 0.5