Anomaly detection aims to identify observations that do not follow the usual behavior of a system. This is important in many applications such as predictive maintenance, fraud detection, health monitoring, and industrial systems.
In many real-world datasets, anomalies are rare, poorly labeled, or not labeled at all. This makes it difficult to train standard supervised learning models.
The method introduces a semi-supervised strategy based on random labeling. The goal is to create a learning signal that helps the model separate normal patterns from unusual ones, even when explicit anomaly labels are limited.
The contribution is to combine deep learning with a simple labeling mechanism that can be used in settings where anomaly labels are scarce or incomplete.
This approach is useful for practical anomaly detection problems, especially when collecting reliable labels is expensive or difficult.