Enhancing Deep Learning Model Privacy Against Membership Inference Attacks Using Privacy-Preserving Oversampling