With timely and proper treatment, an estimated 70% of the 65 million people with epilepsy worldwide could be seizure-free. For the 30-40% of epilepsy patients who do not respond to anti-seizure drugs, detection of pathological interictal activity can speed up treatment procedures. However, analysis of brain activity is a manual process today, which is tedious, time-consuming, and unscalable. An automated approach tackles these issues. That said, interictal detection algorithms rely on supervised learning, which requires labeled training data and human intervention for patient-specific tuning. An unsupervised approach overcomes these limitations with an automated and personalized solution that requires no labels or intervention from medical staff. The algorithm presented in this thesis is a fully unsupervised model that classifies intracranial electroencephalography (iEEG) interictal activity as pathological or physiological. 170 features were extracted from three domains: time (19 features), frequency (7 features), and time-frequency (144 features). Four unsupervised dimensionality reduction techniques were then combined with four unsupervised classification methods. Out of these 16 combinations, Principal Component Analysis (PCA) paired with a K-Means model achieved the best performance. Across 15 epilepsy patients, it yielded an average 92.6% F-2 score, 93.5% precision, and 93.0% recall with standard deviations of 12.0%, 13.8%, and 12.5%, respectively. There was no statistically significant difference between the performance of this unsupervised model and that of two supervised models, Support Vector Machine and Random Forest. The proposed method is a promising approach to enhancing the efficiency of treatment for drug-resistant epilepsy.