Outlier Model Selection

Outlier model selection aims to automatically choose the best-performing outlier detection algorithm and its hyperparameters for a given dataset, a crucial yet under-researched problem. Current research focuses on unsupervised approaches, leveraging techniques like meta-learning, ensembling, and self-supervised learning with hypernetworks to efficiently and effectively select models without labeled anomaly data. These methods address the challenge of inconsistent performance across different datasets and the high dimensionality of hyperparameter spaces in modern deep learning-based outlier detection. Improved outlier model selection promises significant advancements in various fields by enhancing the reliability and efficiency of anomaly detection in diverse applications.