ISSN: 2308-7056 Akhtar & Khalil (2018) 60 I www.irbas.academyirmbr.com August 2018 International Review of Basic and Applied Sciences Vol. 6 Issue.8 R B A S Link Prediction Techniques in Complex Networks M. USMAN AKHTAR University of Engineering and Technology, Peshawar, Pakistan Email: ua@uetpeshawar.edu.pk M. IMRAN KHAN KHALIL University of Engineering and Technology, Peshawar, Pakistan Abstract In this paper we discuss how recently emerged machine learning approach, and conventional graph theoretic approaches used for the prediction of missing links in real world complex networks. Future projected plan can be build based on prediction results. This paper shows that the machine learning approach is significantly good as a newly emerged field. If any real-world situation can be mapped in to complex graph. where the nodes in the graph represent different objects of real world, and the links in the graph denotes the link, then a subset of these links is given as an input to the algorithm for prediction. It will also be describe the mechanism of different structural based link prediction algorithms. Keywords: Graph Theoretic Approach, Machine Learning, Complex Networks, Neural Networks, Link Prediction. Introduction Due to the evolving nature of world real world complex networks also evolve over time, new links and edges are added (Dorogovtsev & Mendes, 2002). The structural based link prediction approaches such as graph theoretic and machine learning link prediction approach uses past data to predict the future structure of a complex network (Zhu & Xia, 2015). Define the links that may appear in the near future and yet not part of the complex network. The missing link prediction have variety of uses in different areas generally used in online social networks such as Linkeidn, google plus, twitter, facebook new friends (Liben‐Nowell & Kleinberg, 2007)and events are recommended on the bases of different parameters likes event or friend suggestion on the bases of common friends and event occurring near residence. Utilization of missing link prediction can also be observed in recommender systems (Schafer, Konstan, & Riedl, 2001), such as new items on daraz online store is recommended on the bases of previous grocery shopping list, in the same way our user previously requests related books and movies are recommended on amazon online book store and netflex an online movie store. Link prediction also used to predict medicine formulation for any newly appeared fatal health care problem. Conventional methods based on the similarity score for missing link between pair of nodes, such as nodes degree, common neighbors, shortest distance between pair of nodes, where a missing link score compared against threshold value for the missing links and edges. with the highest scores are considered the most likely edges. In this paper, we present different structural based link prediction algorithms with supervised learning algorithms for the link prediction problem. Current network state used as an initial input to train a learning algorithm to predict links which may appear in the near future but are absent in the input network. Then, for each missing link, we use the threshold value of the learning algorithm as our link prediction heuristic.