This study proposes a nonlinear data-driven method to delineate Electroencephalogram (EEG) recordings as either coming from controls or patients with epilepsy. This method uses the probability of recurrence and the correlation between electrodes to extract the phase synchronization and the functional connectivity maps of the brain from interictal EEG data recordings. This newly proposed algorithm utilizes probabilistic clustering by extracting graph theoretical features from the calculated functional connectivity matrices. Results reveal that brain connectivity networks of epileptic and control populations show statistically significant differences (t (340) = -37.4771, p<;0.01) between them. Performance results show an accuracy of 92.8% with a sensitivity of 85.7% and a specificity of 100%, when tested on 14 subjects. These preliminary results confirm that this method can be used to enhance and validate diagnosis of epileptic patients from controls using non-invasive scalp EEG signals.