Data Sciences, Machine Learning And AI: Identifying The Need
I write about fintech, data, and everything around it
As we were demonstrating a data anomaly solution for our client using “R” and “Python” today, my thoughts went back to a recent article that I read titled, “India’s mess of complexity is just what AI needs” written by Varun Aggarwal, co-founder of Aspiring Minds in the MIT Technology magazine during June 2018.
The reason Varun’s article flashed across my mind was that we were precisely hitting on an important step that the article had highlighted.
The article says,
“The first crucial step in improving efficiency through robotics and AI is identifying a business problem and converting it to a data sciences problem.”
Artificial Intelligence In Banking
There is a general tendency to apply data sciences for everything today. A number of problems can actually be handled without Machine learning/Deep learning but given the “mouth talk” about AI in the market, everyone wants to put together a solution that uses TensorFlow + Python and claim that the problem has been addressed using AI. Let me give a glimpse of the solution we shared with our client today and how we identified the business problem before determining the course of action (oh and by the way, this is not a machine learning problem). The most important value that you offer when you know what AI is, not recommending it when it is not needed. We remember why we first went into this business (to make our customer’s life simple).
Our client has a trading platform for power generation/utility companies. The platform offers people to buy and sell power. So, they would be taking how much power that gets generated (in load megawatt hours) from each of the sources and forecast the kind of pricing that would be a reference point for subsequent trading. And on occasions, there could be false positives being pushed from energy generation sources that could be anomalies. Such anomalies when they are high in frequency or at extreme values could significantly change the basis of trading.
Our client wanted to have a means by which they can find anomalies and exclude them from the distribution for further computation. Considering the large and varied sets of data involved, it would be a good itch for AI to scratch, however, after much deliberation decided to keep the solution simple and subsequently try out ‘cooler’ stuff later after we taste initial success. The reason to walk from AI treatment was to:
- Keep the technical-debt under control. Don’t dish out solutions that require higher intellectual property for maintenance.
- R as a platform already has many packages meant for Anomaly detection and our customer has existing production deployment ‘R’ based solutions.
- The problem will get a fair treatment even with numerical analysis tools even though the business problem has the potential for further refinement and betterment using neural networks /deep learning which apparently is on the back of our minds.
The key element, however, was to highlight the reasoning behind applying data sciences in a scenario where the use was imminent.
As Varun says,
The IT industry still requires people to write programs. But even there, automation is playing a role in services beyond hard-core programmings, such as network monitoring, testing, and infrastructure maintenance. The big opportunity for the Indian IT industry is to provide machine learning services to the world. IT companies have started to build AI practices, but the country lacks trained talent.”
Just in case you are interested, you can read the full article here.
Related Posts
