• Computer Science , Data Science , Higher Level Management ,
• 12 Jan

When you start learning about the nuts and bolts of Machine Learning (ML) or Artificial Intelligence (AI), there are two concepts that will quickly surface to which you’ll need to understand. They are pretty basic ideas but a lot of ML and AI is understanding how to address a problem and this is part of that.

Supervised Machine Learning

Supervised means that we know the categories or buckets that our data will fit into. These algorithm’s play a matching game of assigning an input to an outcome.  We take historical data and use it to train and test our models, then fine tune for performance. Most of ML and AI fit into this type of learning.  The three main ways this is done is with regression, classification, and then a hybrid style analysis.

In the Iris dataset, which is often used as the “hello world”  of ML, you’ll see that different attributes of the flower fit them into different classes. The part that makes it supervised learning is that we know the classes. We can use what we know about the different classes  or “types” of iris and match the one we find to a known class. This is how we can use supervised machine learning to identify the  “type” of iris by comparing it’s features to that of the features of the iris types we already know exist.

Unsupervised Machine Learning

Unsupervised means that we don’t know the categories or buckets. Our inputs don’t have known outcomes to match them to. This is data discovery as it’s up to the algorithm to look for structure within the data. This often involves clustering points into new categories, looking for relationships,  causes and searching for hidden patterns. We do not have a presupposition about how the data works together with unsupervised.

This can happen in a number of situations such as customer segmenting or correlation of behaviors in individuals. Correlating behaviors may be something like, people who typically buy this product also buys that product or prefers these types of smells, price points, contact points etc.  Keep in mind that correlation does not equal causation.