Remember the days when we thought that no computer can ever master the ancient Chinese game ‘Go’?
Remember the days (2000s) when we thought if at all a computer program has to be designed to win Go, it’ll take at-least 3 decades?
And remember how suddenly one day Google DeepMind’s ‘AlphaGo’ defeated one of the top Go players Lee Sedol and subsequently world champion Ke Jie?
Similarly, (but a very long time ago) IBM’s Deep Blue had beaten the then world champion Garry Kasparov at chess in 1997.
It’s amazing to see how something that we saw as a future, is present in our ‘now’. And it’s even more interesting to know how.
What is machine learning
The technology that enables computers/machines, to learn from the world, acquire knowledge of humans and put it to application, is called machine learning. Artificial Intelligence’s quest to achieve human-level intelligence and go beyond has been one of the driving forces behind these advancements.
History of machine learning
There were some long standing problems when it came to machines. The classic example is- the problem of reading and recognizing optical characters namely text and/or images. Machines couldn’t do this task although they are very natural to humans and most other living organisms. But now, we have machines/ML algorithms that recognize not just these, but also your face plus biometrics and use this information to not only unlock systems but also model complex behaviours.
Machine learning in Credit/lending
Currently, majority of lenders (banks and NBFCs) make lending decisions in ways that have been built over years of analysis on specific types of data, particularly the banking data. They further call for the borrower’s credit-score from the bureau, and the decision is made. Today we have the ability to collect much more data, such as alternate data, about individuals and groups than earlier times, which opens up new avenues for credit scoring and lending in general.
With such novel high dimensional data, as with many other practical domains of application, machine learning algorithms are helping to learn much more accurate models than ever before about credit behaviour of the people.
Machine learning based decision-engines
At CreditVidya, we’re working on building various decision-engines. We have built products that are data-backed and have the capacity to solve complex lending decisions. And we do this by using the large amount of alternate data that we have on hand, and the skill set of big data and artificial intelligence.
Once the features are built using the vast data pool we have, we employ ML methods to consider various possible combinations to decide whether or not the person will default. This is done by correlating a person’s behavioural and psychological patterns with that of their credit behaviour, thus, building highly accurate prediction-based risk models. Furthermore, the timelines of various users also help in building very compelling cross-sell/upsell models.
Do these models work?
We understand that testing, monitoring and revisiting are a very important part of any process, which is why we set up our test-bed and feedback mechanisms very carefully. We prudently picked our test cases- the ones that replicates real-life situations where it is going to be used.
Often, understanding the models learnt by the machine is done retrospectively esp. since the models/neural networks get tuned and optimized faster than a human’s capacity to analyze. What’s more important than the decision given by the machine is, what variables it used, what variables were given more importance while arriving at that decision and whether the final model is explicable enough to go live.
To have a first-hand understanding of our machine learning and artificial intelligence based products, request for a LIVE demo here.