Building Highly Accurate Fraud Detection System

As part of the ongoing interview series at CTOtalk, we spoke to Vijayabalan Balakrishnan, Senior Cloud Architect at the European Commission about his 21 year journey in designing and deploying business solutions in the cloud.

A cancer-survivor himself, Vijayabalan has worked with children affected by cancer and used tech to make a difference in their lives. He founded CloudnLoud in 2011, and through his training and revenue-sharing model with hospitals, has touched the lives of 6000+ children.

Here are snippets from the interview:

The early years

A diploma holder, with no formal college degree, Vijayabalan has relied on hands-on learning rather than academic knowledge. He started as a computer cleaner in his early stages and was later fascinated by the concept of ‘reengineering’ and started modifying existing components and creating custom-made solutions. Vijayabalan has faced numerous challenges along his professional journey. He faced multiple rejections by organizations he had applied at, due to his medical problems and the social stigma of cancer.

But after all odds he joined IBM. Meanwhile, he also began organising town-halls to educate people on how to learn and use technology to their full potential. He also set up his own lab to study re-engineering. He moved to Europe and joined the European Commission in January 2020 and is now leading a large team who work on Data and AI, enabling robust digitalisation across Europe.

“Attempting and experiencing each version of the technology helps us build skills stronger even after multiple versions,” says Vijayabalan.

Project Re-engineered

Vijayabalan recently re-engineered a project using KNIME for fraud detection in digital payments. KNIME allows users to visually create dataflows and selectively executes some of the analysis steps and later inspects the result in order to avoid granular flaws. KNIME allows to create predefined models according to different needs of the governance. Through this he created a model where any client can fit their environment where the clients needn’t have huge investments but adapt this model and train their data to fetch more frauds in their deliverables.

All about KNIME

KNIME is a product that serves the data analytics space. Basically, one can have more than one problem to solve through KNIME using custom-based fraud detection.

Fraud Detection techniques mostly stem from anomaly deduction (branch of data science). We have an enormous amount of data but to analyse the data to come with required results, it’s quite difficult, perhaps could take several years. But products like KNIME can help achieve this faster. If a dataset has a sufficient number of fraud examples, it is enough to create a one single sample and then the same sample can be used for other examples. The models alone can be changed and re-fed into the system every time. This supervises machine learning algorithms to classify random forests. KNIME regresses continuous models to find out all the frauds which cannot be done manually. If the dataset has any fraud examples, you can define what kind of a predicted fraud.So the KNIME will create those algorithms that can be simulated in a multiple set of data either outlier detection approach or isolation forest technique.

Another area where KNIME can be used is, if a new customer has a new segment of data, they can adapt to this existing model which is very easy. This may not require them to have a huge expertise except for basic data understanding.

Basically, you can feed your data and the model gets trained according to the data and the model can be used for their system. Once in a while KNIME shows alerts when it’s not able to model the fed data and keeps track of it. It analyses both fraudulent and non-fraudulent data sets and applies its created model based on its predefined model. Vijayabalan had recently executed this adaptation in a couple of major financial firms and received good results & feedback but of course with a couple of challenges and working on solving them and hoping in a year this model will be even more efficient and successful. The solution automates the process of fraud detection through the ability of tailored ML algorithms. They have created a specific functionality with more detailed algorithms to learn from historical fraud patterns and recognise them. This will help most of the clients in ease of digitalisation.

KNIME in action

One of the most difficult and challenging in the implementation phase is detecting frauds followed by increasing accuracy to prevent the embracement of customer chords i.e., being declined when making a genuine transaction at a store. So he built a very simple approach using label encoding for categorical variables, filling all missing values with 0 and a random forest without any tuning or methods to handle the imbalance in data. Using this tailored approach if anyone has the fast historical data, then can easily train their own model right away.

Of course, KNIME is a very huge feature product but what he developed is used specifically for digital payments in their business. For businesses who want to scrutinise digital frauds in big time such as banking and e-commerce platforms can use the KNIME’s tailored approach he had created.

About CloudnLoud

CloudnLoud is an NGO started by Vijayabalan with an aim to help cancer children who are not able to afford the treatment cost. CloudnLoud has tie up with around 60 hospitals in India and the revenue received as fees through these training programs are used for treating cancer children. Whenever there is an enquiry or request they scrutinise and wet the information of the patient and release funds to the hospitals. He is also inviting more folks to join in as volunteers. He is also supporting 34 product companies as a bug fixing partner and generates revenue which is directly redirected to Cloudnloud funds.

CloudnLoud is now planning to launch an application workflow which allows people to directly apply here for financial support. These requests will be verified by the NGO team and necessary actions will be taken.

Vijayabalan always insists “Trying is not a Crime” via his training .

Watch the full video of Vijayabalan Balakrishnan’s Interview here.