In the last decade, online social networks showed enormous growth. With the rise of these networks and the consequent availability of wealth social network data, Social Network Analysis (SNA) led researchers to get the opportunity to access, analyse and mine the social behaviour of millions of people, explore the way they communicate and exchange information. Despite the growing interest in analysing social networks, there are some challenges and implications accompanying the analysis and mining of these networks. For example, dealing with large-scale and evolving networks is not yet an easy task and still requires a new mining solution. In addition, finding communities within these networks is a challenging task and could open opportunities to see how people behave in groups on a large scale. Also, the challenge of validating and optimizing communities without knowing in advance the structure of the network due to the lack of ground truth is yet another challenging barrier for validating the meaningfulness of the resulting communities. In this thesis, we started by providing an overview of the necessary background and key concepts required in the area of social networks analysis. Our main focus is to provide solutions to tackle the key challenges in this area. For doing so, first, we introduce a predictive technique to help in the prediction of the execution time of the analysis tasks for evolving networks through employing predictive modeling techniques to the problem of evolving and large-scale networks. Second, we study the performance of existing community detection approaches to derive high quality community structure using a real email network through analysing the exchange of emails and exploring community dynamics. The aim is to study the community behavioral patterns and evaluate their quality within an actual network. Finally, we propose an ensemble technique for deriving communities using a rich internal enterprise real network in IBM that reflects real collaborations and communications between employees. The technique aims to improve the community detection process through the fusion of different algorithms.