How Does Machine Learning Work?
Machine learning is a groundbreaking technology for solving problems with computers. But how does it actually work? In this article, you’ll learn what machine learning is and how it differs from traditional programs and IT systems.
Digitising Processes
Machine learning often comes into play when organisations want to digitalise processes and decision-making. Let’s look at an example to see how this might work in practice.
Imagine a bank that offers mortgage loans. In this bank, every application is reviewed manually by a human before an interest rate is proposed. Customers find the process too slow, so the bank wants to use computers to speed things up. How could they do that?
Rule-Based Programming
One approach is to use traditional computer programs. In such systems, data is processed using predefined procedures and rules—hence the term rule-based programming. In the bank’s case, this would mean identifying all relevant factors in a loan application and defining how they should be combined to calculate an interest rate.
So, what’s the limitation of rule-based programming? The challenge lies in defining all the rules. The relationships between different factors can be highly complex. It’s difficult to identify and correctly handle all possible exceptions and edge cases. Rule-based systems follow instructions strictly—if the rules aren’t perfect, the results won’t be reliable either.
Machine Learning
An alternative solution is machine learning. With machine learning, data and mathematical models are used to identify patterns and make predictions. If the bank chooses to use machine learning, it could, for example:
Collect data from previous interest rate decisions, case by case
Build a mathematical model that can calculate an interest rate based on all parameters
Adapt the model to the data—a process known as training
Evaluate the model
Use it in production if it performs well enough
Unlike rule-based programming, you don’t need to explicitly tell a machine learning model how to solve the task. All you need to provide is data—the training process then identifies the patterns required for your specific use case. This shift in focus from manual rules to data is what makes machine learning truly transformative.
Which Method Should I Choose?
Different problems call for different solutions.
Rule-based programming works well when the process involves few and clearly defined rules.
Machine learning is often the right choice when relationships are too complex to be described with simple rules and a data-driven approach is required.
It’s important to remember that machine learning is not always the best solution. Training a model requires high-quality data, preferably in large volumes—and that data isn’t always available. In simpler cases where rule-based programming is sufficient, a machine learning model can be unnecessarily complex to develop and maintain. A careful evaluation based on your specific situation is always essential.
