In this paper, we proposed and evaluated a new network link prediction method that can be used to predict missing links in a social network. In the proposed model, to improve the prediction accuracy, the network link prediction problem is transformed to a general object-object match prediction problem, in which the nodes of a network are regarded as objects and the neighbors of a node are regarded as the node's associated features. Also a machine learning framework is devised for the systematic prediction. We compare the prediction accuracy of the proposed method with existing network link prediction methods using well-known network datasets such as a scientific co-authorship network, an e-mail communication network, and a product co-purchasing network. The results showed that the proposed approach made a significant improvement in all three networks. Also it reveals that considering the neighbor's neighbors are critical to improve the prediction accuracy.