Context Discovery for Anomaly Detection

Contextual anomaly detection aims at identifying objects that are anomalous only within specific contexts. Most existing methods are limited to a single context defined by user-specified features. While identifying the right context is not trivial in practice, there is often more than just one context in real-world systems under which different anomalies naturally occur. In this work, we introduce ConQuest, a new unsupervised contextual anomaly detection approach that automatically discovers and incorporates multiple contexts useful for revealing contextual anomalies. In ConQuest, we search for relevant contexts by optimizing an unsupervised multi-objective function, where each objective is derived from desired properties of contextual anomaly detection. To effectively balance such (often competing) properties, we use a multi-objective genetic algorithm that returns a Pareto front comprising diverse, non-dominated solutions. Through experiments on various datasets, we show ConQuest outperforms state-of-the-art methods. Further, we showcase the advantage of using multiple objectives over single-objective context discovery strategies and demonstrate the interpretability aspect of ConQuest.Som manuskript i avhandling / As manuscript in thesis