Improving Imbalanced Machine Learning with Neighborhood-Informed Synthetic Sample Placement