Detecting money laundering in transaction monitoring using hidden Markov model

Abstract

The purpose of the thesis is to introduce, build and test HMM as a method of detecting suspicious financial transactions that might be correlated with money laundering. HMM is a statistical Markov model in which the system being modelled is assumed to be Markov process with unobserved (i.e., hidden) states. These hidden states however generate observable outcomes. HMM fits the context of transaction monitoring in the fight against money laundering as the intent of a transaction (part of money laundering scheme or not) is and only some parameters of the transaction can be observed. The model was built and tested on artificial datasets provided by Salv Technologies and commonly used k-means clustering model was chosen for comparison. Analysis and testing showed that overall, HMM outperforms k-means clustering. Based on analysis, it can be concluded that in essence, HMM can be used in transaction monitoring but getting high precision needs expert knowledge and practical testing. A brief overview of money laundering, anomaly detection methods and HMM are given. Empirical part includes application of HMM on 3 different study cases using R software.