As we all know, the very popular and widely used concept “big data” has very specific characteristics which differ from the traditional data; such as unstructured format, data volume exceeding 1 PB (Petabyte) and also continuous data flow. As this data is really very big, analyzing it also requires more sophisticated techniques.
In that respect, data mining and machine learning which are based on exploration and analysis of large quantity of data, in order to discover meaningful patterns and also to learn from that data and make predictions, started to be used for big data analytics. Especially, in the financial sector, you can see the use of data mining and machine learning for various processes such as marketing, CRM, risk management, fraud detection and others.
In our case, we also felt the necessity of having an advanced fraud detection tool, using data mining and machine learning algorithms since rule-based traditional data analytics were not enough to detect the internal frauds in an environment where everyday millions of transactions are processed and digital banking is growing rapidly. Mainly, rule-based data analytics resulted in too many false-positive results and was much time consuming.
Then, together with our IT Department, we launched a project in order to develop an in-house fraud detection tool, which will enable us with the following features;
- Having an easy interface, integrated with Core Banking Platform
- Composing scenarios or definitions without any IT development
- Decreasing false positive results and making a more accurate sampling
- Analyzing links and relationships
- Making quick queries for past transactions
Project took about 2 years and completed successfully. Fraud detection tool has been built upon data mining and machine learning algorithms. The model behind the tool is continuously fed by the defined datamarts and trying to detect anomalies within this data in order to reach the “fraud risk score” for each transaction. Shortly, based on all these data analysis, the model is assigning a “fraud risk score” to each transaction, in order to select the riskiest sampling for further investigation. The transactions with the highest scores are chosen by the data analytics team and examined thoroughly.
The model has been developed by powerful tools and uses the most efficient techniques. All past fraud cases have been loaded into the model for learning the patterns and predicting the similar ones. Model also learns from the new detected fraud schemes and adapts itself. Furthermore, “employee risk matrix” module of the tool is still under construction and mainly, this module will try to detect the riskiest employees (in terms of fraud risk), again by getting data automatically from different sources and using a model for prediction.
In my presentation, you will be able to hear all details about the use of machine learning in financial sector, the tool and the model and its practical use
Ramazan Isik (Chief Audit Executive / EVP, Member of Executive Committee, DenizBank) will speak on Use Of Data Mining And Machine Learning For Fraud Detection at the 10th Annual Internal Audit & Governance, Risk and Compliance Forum. Click on the link below, to know more about the conference: