High frequency oscillations (HFOs) during inter-ictal state have been considered as a potential biomarker of epileptogenic regions in brain. The purpose of the current study is to improve and automatize the detection of HFOs basing on HFO distinguishing features followed by unsupervised clustering method, and to predict seizure onset zone (SOZ) using the clustered HFOs. The algorithm successfully separated HFOs of different sub-categories from noise, artifacts, and inter-ictal spikes. We tested this technique on two subjects, and assessed the performance of SOZ prediction by computing the overlapping rate of HFO generative channels and seizure onset channels. In both subjects, we were able to localize the seizure onset area 3 to 4 days before the actual onset of the seizure, with high specificity over 95%. The algorithm showed significant improvement comparing to another existing technique.