AIDA: Analytic isolation and distance-based anomaly detection algorithm

Many unsupervised anomaly detection algorithms rely on the concept of nearest neighbours to compute scores. Such are popular because there no assumptions about data, making them a robust choice for unstructured datasets. However, number (k) neighbours, which critically affects model performance, cannot be tuned in an setting. Hence, we propose new and parameter-free Analytic Isolation Distance-based Anomaly (AIDA) algorithm, that combines metrics distance with isolation. Based AIDA, also introduce Tempered Isolation-based eXplanation (TIX) identifies most relevant features characterizing outlier, even large multi-dimensional datasets, improving overall explainability mechanism. Both AIDA TIX thoroughly tested compared state-of-the-art alternatives, proving useful additions existing set tools detection.