Supervised vs Unsupervised Machine Learning
Key Points
- Supervised machine learning uses labeled data to train models that can predict specific outcomes, such as whether factory robots need maintenance (binary classification) or which of several actions are required (multiclass classification).
- Unsupervised machine learning discovers hidden patterns in data without predefined labels, enabling insights when no explicit outcomes are known.
- Regression, another supervised technique, predicts continuous values (e.g., robot temperature) and can be used for monitoring and anomaly detection rather than simple yes/no decisions.
- By feeding real‑time metrics into these models, businesses can automate maintenance scheduling, anticipate equipment failures, and optimize operational decisions.
Sections
Full Transcript
# Supervised vs Unsupervised Machine Learning **Source:** [https://www.youtube.com/watch?v=3fsy2oheRdg](https://www.youtube.com/watch?v=3fsy2oheRdg) **Duration:** 00:06:05 ## Summary - Supervised machine learning uses labeled data to train models that can predict specific outcomes, such as whether factory robots need maintenance (binary classification) or which of several actions are required (multiclass classification). - Unsupervised machine learning discovers hidden patterns in data without predefined labels, enabling insights when no explicit outcomes are known. - Regression, another supervised technique, predicts continuous values (e.g., robot temperature) and can be used for monitoring and anomaly detection rather than simple yes/no decisions. - By feeding real‑time metrics into these models, businesses can automate maintenance scheduling, anticipate equipment failures, and optimize operational decisions. ## Sections - [00:00:00](https://www.youtube.com/watch?v=3fsy2oheRdg&t=0s) **Supervised vs Unsupervised Machine Learning** - The speaker explains how labeled data enables supervised learning for prediction (e.g., robot maintenance) while unsupervised learning discovers hidden patterns without labels. - [00:03:07](https://www.youtube.com/watch?v=3fsy2oheRdg&t=187s) **Untitled Section** - ## Full Transcript
Your business generates mountains of data, but are you really taking advantage of the
insights that could reveal?
You can use machine learning a branch of AI to analyze your data and predict future outcomes
or identify hidden patterns.
Today I'll cover two approaches, namely supervised and unsupervised machine learning.
The big difference between the two is how the training data is labeled.
As the name suggests, supervised learning needs guidance.
We do this by using label data sets.
A label is simply a known value that we specify on each row in the data set.
It could be something as simple as a binary yes or no, a category or a score.
But what if we don't have all the information needed to assign a helpful label?
Well, that's when unsupervised learning comes into play.
With this approach, we use machine learning to detect hidden patterns in data without
our help.
I think a few examples will help illustrate the big differences.
Let's start with supervised machine learning.
Say we run multiple factories, each with robots that need maintenance, but when exactly?
Well, we know from experience that the temperature of our robots and the level of vibration affects
their server schedule.
So we decide to monitor those attributes.
This table contains the information we've collected.
The goal is to get our model to use the data to predict the label maintenance needed, yes
or no.
And basically, that's supervised learning.
We provide labeled data, and supervised learning produces a model which can accurately predict
that label.
Now, when the prediction is something like yes or no maintenance needed or not needed,
we call that binary classification.
Then there's multiclass classification.
It's a bit of a tongue twister that indicates any number of states.
So in our robot example, a multiclass classification might indicate if the robot needs maintenance
or not, needs replacement, or just need some rest.
In other words, the multiclass approach provides more context using our binary classification
model.
We were able to schedule maintenance on demand by feeding real time metrics into our model
and getting a simple yes or no using the multiclass classification model.
We got that and we added whether we might need a replacement robot or could just give
a robot a break.
Another popular form of supervised machine learning is called regression.
Regression is used when you want to predict a continuous value like temperature.
Back to our factory.
We can train a model to predict what a robot's temperature should be, given how hot it is
in the factory and the robot's power consumption.
With that model, we can compare the predicted temperature to the robot's actual temperature.
Unlike our first example where we predicted maintenance needed or not with regression,
we might use a dashboard to keep an eye on the situation.
Instead of getting a predictive yes or no.
Okay, now we understand the basics of supervised machine learning.
Let's take a look at unsupervised machine learning.
Remember, this is the approach that learns from a data set without labels.
Instead, we look for patterns in the data.
There are three main types of unsupervised learning.
The first is called clustering, which groups unlabeled data based on similar characteristics.
For example, an online store might use clustering to develop customer personas.
That group people with similar buying patterns.
So how is clustering helpful?
Well, take a look at this dataset.
Even though this is relatively small, it's still hard for the human eye to detect any
patterns.
Now, look at this image we produce with clustering in this 3D diagram.
We've plotted out the numbers in that table.
Now we can easily see clusters or spot any outliers.
This has major implications for online retailers.
Using this model you can tailor your user experience for each group.
You can also flag customer behavior that doesn't fit any persona and could be identity fraud.
The next type of unsupervised machine learning is called association.
Association is used to identify relationships in the data.
Picture your music streaming playlist when your platform pops up a message.
Listeners who liked X also liked Y that's association.
Finally, there's dimensionality reduction.
It helps eliminate noisy, redundant data from unmanageable datasets.
This reduction simplifies the input data before training a model.
It's sort of a smart, tactical way to trim fat and actually get more accurate results
with less data.
Okay, so that's my brief overview of supervised versus unsupervised machine learning, but
the question remains, which approach is right for you?
If you want to predict outcomes and you're willing to train your model by manually labeling
your data, supervised learning is your best bet.
In our factory example, knowing what robots needed servicing reduced downtime with proactive
and cost effective maintenance, but if you have lots of data and are struggling to identify
patterns, go with unsupervised learning.
It can help clear the fog that's preventing you from seeing what the data is telling you.
In our simple music shopping example, grouping shared interests helped improve the customer
experience and boosted sales.
When you're ready to start, think of as an experiment.
There are many different tools you can try.
Check out the links in the description to get started.
If you want more information on machine learning and other technology topics, check out the
free resources available to you on developer.ibm.com If you like this video and you want to see
more, please like and subscribe.
If you have any questions, drop them in the comments below.
[00:00:00][0.0]