The infrequent nature of seizure events in epilepsy has engaged researchers to find alternative ways to accurately predict seizure onset regions. We implemented a hybrid approach combining the granger causality based network estimation with subsequent measurement of graph centrality measures in the interictal electrocorticography (ECoG) data to predict seizure onset zones. Critical to the success of this method was dividing analysis into two stages: first, detecting epileptic spikes and second, identifying their functional network. We analyzed interictal data from 8 epileptic patients and found that the epileptic spike functional network is a robust predictor of physician-identified seizure onset compared to the commonly used total ECoG network. Moreover, we discovered that PageRank, indegree and hub centrality measures best predict seizure onset regions. This establishes several features for automated detection of seizure onset zones, and can assist physicians in surgical planning and decisions.