Fraud detection model based on the discovery symbolic classification rules extracted from a neural network

Abstract

This paper presents a fraud detection model using data mining techniques such as neural networks and symbolic extraction of classification rules from trained neural network. The neural network is first trained to achieve an accuracy rate, the activation of the values in the hidden layers of the neural network is analyzed and from this analysis are generated classification rules. The proposed approach was tested on a set of data from a Colombian organization for the sending and payment of remittances, in order to identify patterns associated with fraud detection. Similarly the results of the techniques used in the model were compared with other mining techniques such as Decision Trees and Naive Bayes. A prototype software was developed to test the model, which was integrated into RapidMiner tool, which can be used as a tool for academic software. © 2010 Springer-Verlag.