Financial fraud detection : a case in anti-money laundering

Abstract

The main aim of this dissertation is to attempt to identify transactions which are fraudulent. Fraudulent transactions are taken to be transactions that are attempting to perform money-laundering. Two classification techniques have been identified to identify these transactions. The first is the Support Vector Machine. This technique is a supervised machine learning algorithm attempts to fit a maximal margin hyperplane, often referred to as the decision boundary. The next classification technique used is the Relevance Vector Machine. This technique uses a Bayesian Framework to classify transactions. This was a method proposed to address the limitations of the Support Vector Machine. The data used was a synthetic financial dataset called PaySim. PaySim is a mobile money transaction simulator which was specifically developed to address the lack of datasets containing financial fraud. A 4-fold Cross-Validation was used for data analysis.