Secure Payment Terminal

Inspiration

Thanks to Visa contactless payment. Visa’s payWave offers convenience, efficiency and multiple layers of security to the card user. The credit card user will need to check the amount, wave, and tap to go.

However, credit card fraud has become more common in recent years. Hackers can use a variety of methods to gain illegal access to the credit card information, and perform unauthorised purchases online. While EMV (embedded on credit card) helps to encrypt the users’ information, the information accessible to the hacker is very minimal and difficult to make unauthorised transactions. Nevertheless, if the credit card is handled carelessly (for example illegal access to the physical credit card), the thief could make contactless payment at the physical terminal.

Our proposal works on the concept of improving security when performing physical transactions. We proposed facial verification and limit the transaction based on past transaction records.

What it does

One of the solutions my team has come up with is AI Face recognition when making a physical transaction. This AI recognition will include facial recognition (or biometric metric) to increase security while offering convenience to the shoppers.

Second solution is to provide a transaction limit based on the user's spending behaviour. This machine learning works based on the past transaction record to set the transaction limit. If the fraud occurs with the transacted amount exceeding the limit, the transaction will be terminated immediately.

How we built it

My team will implement a camera on the payment terminal where we will capture the user’s face. The AI system will perform verification if the transaction is performed by the card owner. If fraud use of a credit card is identified, the transaction will be disallowed.

We will also use the data analytic concept to perform data cleaning and data aggregation. If the user dataset is large enough, clustering algorithm will be performed to cluster different timestamps of the transactions. However, in our prototype, due to insufficient data, we will cluster the timestamp based on the most frequent transaction. After clustering, we will set the max limit according to the mean + 2* std deviation to encompass 95.4% of previous transaction activities

Challenges we ran into

While brainstorming, we ran into many challenges like how user-friendly the AI system was, what issues we will face when implementing this system. We have tried VGG16 and DeepFace framework to test on members’ photo to perform facial verification. We keep trying on various hyperparameters to optimize the verification process. The time it takes to verify the user’s identity by using a system (on non-GPU environment) is approximated at ~3.0s. The cosine similarity works well under the epsilon threshold of 0.4.

To have a better understanding of user purchasing behaviour, we were lack of the user transaction record. As such, we extracted the transaction record contributed by one of our team members. In future, we will get sufficient data for better clustering of the spending behaviour.

Accomplishments that we're proud of

We have finished one GUI which allows the user to do emulate the verification of user's identity. The system works well with a 100% successful rate. With our machine learning model, the maximum allowable limit of the transaction helps to eliminate the possible fraud use of credit card.

What we learned

Other than AI computer vision and deep learning neural network, we learnt how to create GUI with few button/input/ output features. We are able to shorten the time (~3.0s) to verify the identity of the user. We also learn how to communicate effectively as a team and learn from each other's strengths and weaknesses. Through this process, we learn the importance of teamwork.

What's next for Secure Payment Terminal

Secure Payment Terminal features will be integrated into existing online banking systems and Automated Teller Machine. The owner's mobile gadget location must match the card usage location. Machine Learning will be used to track user spending behaviour and notify the user if fraud use of a credit card is identified.

Add-on features like biometric thumbprint and continuous learning will be performed by the inbuilt system.