QID Project

Independent Component Analysis (ICA)

ICA is a statistical technique for decomposing a complex dataset into sub-parts that are as independent from each other as possible. Recently, ICA algorithms have been used to reveal hidden factors that underlie sets of random variables, measurements or signals.

The premise of this project is that valid access and fraudulent access are statistically independent patterns of activities.

The research will investigate the following issues required to apply ICA to the problem of fraudulent authentication:

• Design of multi-dimensional model for user behaviour after successful authentication.
• Methods for pre-processing the data model prior to applying an ICA algorithm.
• ICA algorithms applicable to the problem of fraudulent authentication.
• Methods for identifying rejection of valid users (Type I statistical errors).
• Methods for identifying authentication of invalid users (Type II statistical errors).