DIGITAL WALLET FRAUD IDENTIFICATION
Finance – Anti Money Laundering
Anti-money laundering refers to a set of laws, regulations, and procedures intended to prevent criminals from disguising illegally obtained funds as legitimate income. Though anti-money-laundering (AML) laws cover a relatively limited range of transactions and criminal behaviours, their implications are far-reaching.
Up to now, most of the banks used a rules-based detection approach to AML and are gradually moving away from a rule-based method to an AI approach. An AI driven system that can gather data instantly and be programmed to make decisions when given a set of simple facts improves the AML layer, reduces false positives and hence results in cost saving.
Solution Overview
We analysed customer transaction data for 6 months to build a model that can predict whether the transaction is fraudulent or not. The various factors that influenced the transaction being fraudulent or not were Credit by Debit Ratio, Transaction Amount , Transaction Time, New Balance Sender, New Balance Receiver, Total Amount transferred by a customer in day, High Risk Geography, Education etc.
Quote To Cash
NIIT-Tech built a modelling system to integrate the classification models, which helps the Banking and Financial institutions to reduce the false positive rates for AML cases and visualize a true 360 degree view of the customers of the Bank.
Know Your Customer
The system provides Know Your Customer snapshot based on the customer’s data about their demographics. Moreover, the transaction details and insights of the customer is readily available enabling better business context in terms of Anti-money Laundering cases for the AML investigators.
Reducing False Positive rates
Through deep analytics and applied models done on a bank’s data, risk and compliance can be made more decisive thereby driving down the false positive rates which at the moment is the cause of concern for the banks. Banks are experiencing a false positive rate of about 95-99 percent which implies that only between 1-5 percent of all alerts result in an actual filing of a Suspicious Activity Report (SAR).
The system provides a drill down on the sender and beneficiary relationship which can provide insights into the tracks of origin and final destination of the money under circulation.
Geographical Analysis
AML investigators can access the system in order to gather information about a particular alert. Through geographical maps the investigators can access the concentration points of fraudulent transactions based on various parameters such as volume and amount of the transactions.
The system provides Know Your Customer snapshot based on the customer’s data about their demographics. Moreover, the transaction details and insights of the customer is readily available enabling better business context in terms of Anti-money Laundering cases for the AML investigators.
Model Details
Algorithms Random Forest, Logistic Regression & RUBOOST classification were used and finally results from Random Forest were utilized, which gave 91.23% accuracy level. Overall difference between actual fraud and predicted fraud was very less, indicating that model has a good prediction capability.
