Class Detection via Anomaly Detection shows really impressive performance. Anybody got an intuition as to why is that so? Something like having opinionated low

Class Detection via Anomaly Detection shows really impressive performance. Anybody got an intuition as to why is that so? Something like having opinionated low-variance high-bias classifiers that you ensemble? In which case, how can it be better than random forests?