The scale of money laundering globally is estimated to be as large as $1 trillion to $2 trillion annually. The overwhelming majority of this money is channeled to organizations trafficking drugs, weapons and human beings, or used to finance terrorist activity. However, despite the fact that almost 70% of that illicit finance flows through legitimate financial institutions, the United Nations Office on Drugs and Crime estimates that less than 1% of the global trade is seized and frozen. In light of that, it's perhaps unsurprising that regulators are stricter today when it comes to enforcement. Since the 2008 financial crisis, regulators, particularly those in the U.S., have been handing down record fines to financial institutions seen to be trading with sanctioned parties and countries, or failing to appropriately comply with anti-money laundering (AML) initiatives.
Who will be the first to implement the new suite of Watson services? From the newly formed Watson Financial Services division, IBM has released the first suite of services covering regulatory requirements, financial crime insights, and financial risk modelling. These cognitive tools have been made possible following IBM's 2016 acquisition of global consulting operation, Promontory Financial Group. Promontory was originally working to provide support to banks dealing with the growing and tightening regulation and risk management within the financial services. It was the knowledge and expertise accessed in this acquisition that brought life to the new financial services-focussed Watson services, with regulation and risk accounting for two thirds of the suite, and a financial crime tool completing the set.
This article describes my machine learning project on credit card fraud. If you are interested in the code, you can find my notebook here. Ever since starting my journey into data science, I have been thinking about ways to use data science for good while generating value at the same time. Thus, when I came across this data set on Kaggle dealing with credit card fraud detection, I was immediately hooked. The data set has 31 features, 28 of which have been anonymized and are labeled V1 through V28.