Abstract
Obsessive-compulsive disorder (OCD) is a chronic and disabling condition affecting approximately 3.5% of the global population, with diagnosis on average delayed by 7.1 years or often confounded with other psychiatric disorders. Advances in electroencephalography (EEG) analysis using machine learning hold promise for the development of OCD-specific biological markers. This systematic review aims to evaluate studies that classify individuals with OCD from other groups based on EEG data. Following PRISMA guidelines, we searched the Web of Science, Scopus, PubMed, and IEEE databases through February 2025; of 42 screened studies, 11 met inclusion criteria for final analysis. Data were extracted across four domains: general information, population characteristics, EEG features, and machine learning features. Results revealed extensive heterogeneity in study populations, associated symptoms, EEG preprocessing methods, validation strategies, and reporting of model accuracy, underscoring the need for harmonized standards. Notably, only a few studies provided statistical interpretation of their models. None of reviewed studies employed modern interpretability techniques such as SHAP or LIME methods that, beyond reducing "black-box" opacity, can inform optimal electrode placement for neurofeedback or transcranial electrical stimulation. Many studies were constrained by cultural limitations, small sample sizes and lack of demographic information e.g., age, gender, medication. This work represents the first systematic review of EEG-ML classification studies in OCD and emphasizes the urgent need for methodological standardization in this emerging field.