This repository contains a Jupyter notebook and a project report related to unsupervised anomaly and novelty detection, using a popular method called the One-class SVM (OSVM). This was done as part of a school project at ENSAE in 2017.

The report contains :

• A brief overview of the theoretical framework behind the algorithm
• A benchmark of its performances on both simulated and real world data (USPS). The corresponding code can be found in the Jupyter notebook.

Main reference :

• Schölkopf B., Platt J. C., Shawe-Taylor J., Smola A. J. et Williamson R. C. “Estimating the support of a high-dimensional distribution” Neural computation, 13(7), 1443-1471 (2001).

Additional references can be found in the report.