In the United States in 2013 alone, credit card fraud cost companies almost $7.1 billion dollars.
Given these enormous costs, fraud detection and classification has become a very active area of research in machine learning and data mining domains.
Although the power of machine learning techniques for fraud detection has greatly increased over the past decades, the incentives for fraudsters to circumvent and adapt to these classification algorithms has also grown.
Effective fraud detection models must be able to adapt to behavioral changes on the part of the adversary, while maintaining high levels of accuracy and low levels of false positives.
The goal of this project is to create a highly scalable and efficient pipeline for detecting fraud in an adversarial environment.
The technical approach will rely on explicit modeling of adversary behavior in a manner that fuses game theory and machine learning.
The basic framework involves adversarial classifier reverse engineering, a relatively new technique for modeling an adversarial environment that does not rely upon the assumption that the adversary has complete knowledge about the classifier.
Within this framework, we will model active experimentation by the fraudsters with the goal of discovering effective fraud attacks that can be found in a reasonable number of queries by the adversary.
Once a set of effective attacks has been found, the adversary would shift the distribution of its transactions to oversample these attacks, thereby increasing its overall effectiveness.
The question then becomes how and when should the lender respond.
With what frequency should the lender rebuild its classifier?