Preventing financial fraud and protecting personal data requires intelligence and self-learning along with computation power. Artificial Intelligence is the key to making a system an efficient e-auditor by removing manual intervention. In some recent cases, the involvement of bank employees in a financial fraud has been amply evident. Artificial Intelligence is needed to reduce the reliance on human auditors without losing the knowledge and self-learning capability of humans.
The system for e-auditing can be made artificially intelligent with the following concepts:
Decision trees are a popular tool in machine learning. The first step in a decision tree is to create a knowledge base. Knowledge base is similar to the memory component of our brain. Knowledge base holds information like – “If a transaction is done from a terrorist country then it is fraud”, “If a transaction involves multiple currencies and the volume of the transaction is high then it could potentially be a fraudulent transaction”. Knowledge base holds information about the parameters and the decision logic. It could start with limited information and data, but the system learns and updates the knowledge as it goes along. Whenever an event occurs the system scans the knowledge base to arrive at a decision. The knowledge base is represented as a tree and the decision is taken by traversing the tree.
Computers recognize data as zero or one. Fuzzy logic is to have a representation between zero and
1. Some of the Fuzzy words that are difficult to represent in computer language are “Beautiful”, “Tall”, “Roast the bread till it turns light brown”. Fraud is also a kind of fuzzy word and we cannot directly identify an event as a “fraud” or “not a fraud”. A huge volume of transaction in a day cannot be considered as a fraud for a corporate customer but it could potentially be a fraud for a retail customer.
So the output from decision trees cannot be used as-is to arrive at the conclusion. Based on the decision from the decision trees a fuzzy set is formed and the events with the highest cardinality are considered as fraud suspects.
To make a system as intelligent as a human brain, we need to borrow some concepts from the functioning of a human brain. Neural network adopts the concept of self-learning and pattern identification from neurons and axons. Human brain remembers and learns the instance based on the impact created by dendrites. It is similar to an impact created by a stone thrown to a clear water. Learning will done based on the communication between neurons through dendrites1.
Once the fuzzy set is created and the fraud suspects are identified the system needs to be sent for learning through neural networks. Human intervention is required to correct the decision and based on the weightage assigned to different parameters the system learns by updating its knowledge base.
To conclude, with the help of AI tools bank lenders can prevent financial fraud and protect personal data.