Anomaly detection using unsupervised method is a challenging one. Isolated Random Forest and Local Outlier Factor are the most promising one. They detect outlier with highest recall possible.
