Clustering algorithms and Autoencoders for anomaly detection

by Ms Alessia Saggio (UCL - CP3)

Friday 3 Mar 2017, 13:00 → 14:00 Europe/Brussels

E/3rd floor-E.349 - Seminar room (E.349) (Marc de Hemptinne (chemin du Cyclotron, 2, Louvain-la-Neuve))

In data mining, anomaly detection (or outlier detection) is a method to identify data points which do not conform to an expected pattern or simply to the other items in a given dataset. In real world, many examples of anomalies can be issued, e.g. bank frauds, network intrusion, errors in a text. Unsupervised Machine Learning techniques are typically used to address the problem. During an internship of three months at B12 Consulting (Louvain-la-neuve, BE), I've been using Clustering techniques and Autoencoders. The former represent a possible method to isolate suspicious behaviors in the data, but showed some weakness (although their validity is still well confirmed in a large variety of contexts); the latter resulted in a promising way to tackle the problem, also providing a quantification of the anomaly detected.

Slides seminar_B12_at_CP3.pdf