Unsupervised Anomaly Detection

The goal of unsupervised anomaly detection is to identify rare objects or events in a dataset that have not been seen before, without any prior knowledge. The core of this task lies in modeling the distribution of normal data and defining a metric based on this model to classify samples as anomalies or normal. Since anomalous data occupy an extremely small proportion in high-dimensional space and are difficult to describe directly, it is usually necessary to map the data to a more suitable feature space for effective detection.