In our first-ever Spanish webinar, we spoke to Héctor Cuesta, an experienced Data Scientist and Director of Product Management at Kueski, to discuss machine learning and its impact on the financial sector. Héctor is also the founder of Dataxios, an AI research company, as well as the author of the book Practical Data Analysis.
What is Machine Learning?
To open the discussion, Héctor directed a pertinent question to the audience.
“How can we make an algorithm that can create generalizations of situations, for something it was not originally programmed for?”
An algorithm usually has a defined number of cases to allow it to function correctly. But, what happens when it faces an unexpected occurrence? That’s where machine learning comes in: with this technology, it’s possible to generalize circumstances to manage and predict any previously unforeseen changes. But, how does it work, exactly?
Three key points in Machine Learning history
Before exploring how machine learning functions, it might be helpful to understand how it came to be in the first place. Héctor points out three crucial moments in its timeline:
- Claude Shannon: in his research paper (1949), the author discusses the Minimax algorithm used by computer systems to predict chess moves. The algorithm takes advantage of the immutability of the basic set of chess rules, to incorporate them into its system, to not only generate moves based on its experience but also to predict those of its opponent.
- IBM Watson wins Jeopardy: At this point in time, machine learning had begun its sophisticated transition to its present form. Besides its ability to predict scenarios based on fixed rules, it now begins to incorporate the possibility of doing so in unstructured environments. This is precisely what occurred in 2011 during Jeopardy, a game show where questions are answered by participants, and where the algorithm was able to learn, implement and predict variable situations.
- AlphaGo vs Lee Sedol: The modern world of machine learning was set in 2016, when AlphaGo – an algorithm created by Google – won at Go, a more complex game in terms of its unpredictability than chess. What is more important in this level of machine learning production is that AlphaGo used simple and accessible resources available to the general public. Despite being easy to obtain, this technology is so efficient that it can even “train” itself.
Exploring what machine learning is shouldn’t be an excruciating task. In fact, the answer could be narrowed down to three basic words: observing, learning and predicting. Nevertheless, the landscape gets more complex and interesting when considering not only the nature of machine learning, but also its applications.
Machine Learning: Why do we need it?
Given the evolution of this technology during the last decade, there are as many subcategories available as there are demands for the technology in the marketplace. Héctor highlights the main three aspects of machine learning:
Supervised algorithms: much like the first example described in this article, this type of approach allows the user to have a database that would be employed to program the algorithm and enable it to predict certain information. Some examples of these kinds of algorithms include those involved in fraud detection.
Unsupervised algorithms: while supervised algorithms work based on a fixed set of values, the unsupervised type uses a system of comparison: this method puts into motion the information it possesses to track similarities in unfamiliar circumstances, which would then be incorporated in its database. This process is normally referred to as “clusterization”.
Reinforcement learning: this sort of algorithm is the result of the previous ones, but with the added value of learning by repetition.
Machine Learning and Fintech
One of the most useful aspects of machine learning is its precision. In the world of fintech, this comes in handy in meeting the requirements of this specialized sector.
“How, precisely”, asks Héctor, “can machine learning be beneficial for fintech?”
This is one of the most valuable assets of machine learning for fintech products: it poses questions that help to predict, attract and filter ideal candidates.
Among the questions machine learning can solve are:
- Which category does this client belong to?
- What is the meaning of this characteristic, in this particular context?
- What characteristic is similar to the original one?
- What will be the value of certain characteristics in the future?
- What is the required outcome?
With these questions integrated into a financial scheme, it would then be possible to form an accurate and custom-made way of approaching clients. For example, knowing in a client’s category would help a financial institution to predict is behavior as a consumer. If a person is in the category of “senior employer” and is also labelled as “financially stable”, representatives could make contact with them to offer a retirement saving plan or similar.
In the same vein, asking “what characteristic is similar to the original” would be helpful to encourage the algorithm to look for, categorize and offer similar or compatible benefits to the one the clients initially obtained. If a customer just received a mortgage approval, a machine learning tool could offer home insurance in addition.
Lastly, this information is then integrated into fin tech-oriented products. These can take countless forms: from tools that help both customers and financial organizations to make better monetary decisions, instruments that prevent fraud and offer security assistance, chatbots that provide a personalized financial experience, and platforms that can evaluate in a single “click” if a client is a suitable candidate for credit or not.
In the ever-evolving world of fintech, the financial sector can greatly benefit from the precision, detailed research and predictions offered by machine learning models.
If you found this blog useful, check out our other blog posts for more essential insights!