Ein-Klassen Klassifikation (OCC) is a specialized Ansatz des maschinellen Lernens designed to identify and classify instances of a single class while treating all other instances as outliers or anomalies. This technique is particularly useful in scenarios where the available data primarily consists of examples from one category, such as Betrugserkennung, medical Diagnosen oder Vorhersagen seltener Ereignisse.
Bei herkömmlichen Klassifikationsaufgaben, algorithms are trained on multiple classes, learning to distinguish between them based on input features. However, in One-Class Classification, the model is trained solely on data from the target class. This method allows the model to learn the characteristics and distribution of the single class, enabling it to recognize instances that belong to this class while flagging those that do not as anomalies.
Gängige Algorithmen, die bei der Ein-Klassen-Klassifikation verwendet werden, sind Support-Vektor-Maschinen (SVMs), which can create a boundary around the target class in the feature space, and neural networks that can be trained to reconstruct input data, identifying deviations from the norm. An important aspect of OCC is its utility in environments where obtaining negative examples (instances not belonging to the target class) is difficult or impossible.
Ein-Klassen-Klassifikation ist ein leistungsstarkes Werkzeug in Anwendungen wie Netzwerksicherheit, where the primary goal may be to identify malicious activity based on normal behavior, or in industrial settings, where monitoring equipment health can prevent costly failures by recognizing deviations from standard operating conditions.