myCADI: my Contextual Anomaly Detection using Isolation - Archive ouverte HAL
Communication Dans Un Congrès Année : 2024

myCADI: my Contextual Anomaly Detection using Isolation

Résumé

myCADI is a machine learning framework associated with a graphical interface for discovering and understanding the internal structure of an unsupervised dataset. It is an intuitive end-user interface to the CADI approach [9], which uses a revised version of the Isolation Forest (IF) method to both 1) identify local anomalies, 2) reconstruct the cluster-based internal structure of the data, and 3) provide end-users with explanations of how anomalies deviate from the found clusters. myCADI takes numerical data as input and is structured around several interfaces, each of which displays a ranked list of the found anomalies, a description of the subspaces in which the different clusters lie, and feature attribution explanations to ease the interpretation of anomalies. These explanations make explicit why a selected point is considered to be a local anomaly of one (or more) cluster(s). The framework also provides dataset and trees visualizations.
Fichier principal
Vignette du fichier
cadix.pdf (5.41 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04743207 , version 1 (18-10-2024)

Licence

Identifiants

Citer

Véronne Yepmo, Grégory Smits. myCADI: my Contextual Anomaly Detection using Isolation. Proceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM ’24), Oct 2024, Boise, ID, United States. ⟨10.1145/3627673.3679208⟩. ⟨hal-04743207⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

More