<p>Imbalanced data, where the sample size of one class (minor class) is much smaller than that of the other class (major class), are frequently met in classification problems such as medical diagnosis, fault detection, fraud detection and so on, and various learning algorithms have been developed for efficient classification based on imbalanced data. When data are extremely imbalanced (i.e. the size of the minor class is extremely small), however, existing classification methods for imbalanced data would not be efficient. Instead, it would be beneficial to treat minor samples as anomalies and apply supervised anomaly detection methods. In this paper, we confirm this conjecture by analyzing various benchmark datasets. Various imbalanced classification methods as well as supervised anomaly detection methods are reviewed first, and then they are compared by analyzing various benchmark datasets as well as simulated data to show that supervised anomaly detection methods are superior to imbalanced classification methods in prediction accuracy. Within our knowledge, this is the first work to compare imbalanced classification methods and supervised anomaly detection methods.</p>