Fig. 3
From: AVC: Selecting discriminative features on basis of AUC by maximizing variable complementarity
Illustration of the Manhattan distance measurement used in AVC. Given that x k is a misclassified instance in the intersection S. Then we can easily find out its nearest neighbors on both features I ik and I jk . (a) shows that I ik and I jk are two separate instances. The red solid line represents the Manhattan distance between two pair of points (x k ,I ik ) and (x k ,I jk ). (b) shows that I ik and I jk are the same instance, so we discard x k in the next calculation