ULfAD – Unsupervised Learning for Anomaly Detection
ULfAD explores unsupervised learning techniques for anomaly detection. The idea is to find patterns of interest such as outliers or exceptions that deviate from normal data behaviour. Nowadays the early detection of these unexpected and rare events is an important part of a business because it helps to reduce the downtime of the processes, reduce money lost preventing equipment damage and allowing make corrective decisions more quickly.
ULfAD is a web-based tool that offers three state-of-the-art methods for anomaly detection based on unsupervised learning schemes in a user-friendly manner:
Three cutting-edge methods: Self-Organising Maps (SOM), Autoencoders and Deep Autoencoder Gaussian Mixture Model (DAGMM).
Two illustrative examples: Intrusion detection prediction on the Internet traffic and Credit Card Fraud Detection.
Individual and comparative analysis of the methods.
Explanation based on intuitive and interactive Charts: Heatmaps, Precision-Recall curves, Confusion Matrices and Boxplots.
The tool is targeted to show to any company or organization the potential of unsupervised learning methods in anomaly detection. The demonstrator presents a series of visualisations and controls that allow understanding the functioning of these methods and the importance of defining properly the controls (thresholds) for reducing according to the case, the false positive or false negative rates depending on business considerations.