While 5G wireless networks are expected to handle the ever growing data avalanche, classical deployment/optimiza-tion approaches such as hyper-dense deployment of base stations or having more bandwidth are cost-inefficient, and are therefore seen as stopgaps. In this regard, context-aware approaches which exploits human predictability, recent advances in storage, edge/cloud computing and big data analytics are needed. In this article, we approach this problem from a proactive caching perspective where gains of cache-enabled base stations in 5G wireless are studied. In particular, huge amount of real data from a telecom operator in Turkey is collected/processed on a big data platform, and an analysis is carried out for content popularity estimation for caching, aiming to improve users' experience in terms of request satisfactions and offloading the backhaul. Subsequently, with this mobile traffic data collected from many base stations within several hours of time interval and the estimation of content popularity via machine learning tools, we investigate the gains of proactive caching via numerical simulations. The results show that proactive caching fulfils 100% of user request satisfaction and offloads 98% of the backhaul, in a setting of 16 base stations with 15.4 Gbyte of storage size (87% of the total catalog size) and 10% of content ratings.