Discriminating Input Variables for Fraud Detection using Radial Basis Function Network

Abstract

Fraud is an adaptive crime; special methods of data gathering and analysis are required to combat fraud issues as criminals often quest for dubious techniques to evade detection. Radial basis function (RBF) network, was used to build base models that identifies and detect the risk of fraud in transactions. At first, it is imperative to isolate the basic factors that are predictive of fraud occurrences so as to determine the Information gain of each attribute. The input variables’ importance was ascertained to indicate how some of the input variables were distinguished as strong indicators or weak indicators of fraud. Hence, the relevant attributes were selected prior to examining the model’s performance. This study has found relevance among corporate business professionals and government agencies, to minimizing the time and cost of fraud detection. The researcher recommended that fraud mining processes be regularly updated at fixed time intervals to checkmate criminals.