The growing demand for machine learning (ML) in recent years has led to an increased need for high-quality resources. ML can add great value to companies, and many sectors are now implementing it to solve complex business problems. They are able to generate actionable insights, assisting with their decision-making, helping them stay ahead of the curve. However, it is important to remember that ML is not the answer to every problem, and is only as good as the data it is given (Garbage in, Garbage out!). When used incorrectly, it can often cause more harm than good.
ML can be overwhelming to people who are not directly working with it, so this blog attempts to bring it back to basics.
ML can have a huge impact on helping businesses identify opportunities such as improving their business efficiency and their markerting reach. There are two very common types of ML; Supervised Learning and Unsupervised Learning, which work on different types of data and solve different types of challenges. Supervised learning is used when you want to build a model that can predict the value of a dependent variable from input variables, whereas unsupervised learning is used if you want to find hidden patterns by splitting data into clusters.
For example, supervised ML can be used to predict whether a customer is going to leave a service or stop buying a particular product. If a business is able to predict which customers are likely to leave, it gives the company a chance to reach out to the customer proactively and offer them an incentive to stay.
Essentially, we can use historic data, to teach the model what kind of customers are leaving a service by giving the model a set of inputs e.g.,
- How long a customer has been subscribed to a service
- The amount of engagement the customer has had with the company recently
The model learns this information (and whether that customer has left or not) to make predictions for future customers.
Unsupervised learning is used to uncover hidden patterns with data. This type of ML works on data that does not have known outputs. A popular technique within unsupervised learning is clustering. This is where we split the data into groups based on some pre-defined similarity; each cluster is made up of data points that are similar to each other in some way. This is, again, popular within marketing, enabling you to cluster groups, where similar people are grouped together. This can allow companies to tailor their marketing approach depending on the customer group.
There is no one-size-fits-all algorithm to solve every problem and it is often trial-and-error, but there are basic considerations that can help guide your decision. Some considerations are as follows:
- Computation time
- Do you need a model that has quick training time or prediction time? or both?
- Explainability and interpretability vs Complexibility
- Do you need a model that can be dissected and explained easily? Or can you use a more complex model if it increases the accuracy but sacrifice on the interpretability?
- Does your model need to handle large volumes of data?
- How is the output going to be used? E.g., are they going to be used to assist a stakeholder to make important business decisions? If so, the explainability may be what they are most interested in. Understanding how the results are generated and how the model is working can provide reassurance that the right decisions are being made.
I have recently delivered a webinar on this subject and you can watch it here. In addition to the webinar, Bays Consulting are offering training courses – Machine Learning Primer. This course will give you hands-on experience in both supervised and unsupervised learning. If you are interested in ML but you are not sure where to start – this course is for you! By the end of the course, you will understand the foundations of ML and will be able build and interpret a simple model in Python. For more details of the course, please contact us on firstname.lastname@example.org.