|
Abstract:
|
With the abundance of information available describing agent based networks, the need to detect fraud and collusion in such networks is paramount. Some approaches have relied on clustering type methods to detect such collusion (Palshikar and Apte, 2008), but such methods lack the capability of quantifying the degree to which a set of individuals may be colluding with one another. Furthermore, since networks are often very large, enumerated searches are quickly untenable, and heuristic approaches must be relied upon. Our research relies on a Genetic Search Algorithm to detect the most likely set of colluders. This genetic algorithm uses, for its fitness function, a likelihood ratio test statistic so that a p-value may be attached to the possible collusion set. In this way, we are able to attach a p-value to claims of possible collusion.
|
