Prediction models (supervised learning)

1. Prediction models (supervised learning)

Welcome to the video on supervised learning models!

2. Supervised vs. Unsupervised

Supervised learning models have a target variable, which can be either a class or type of an outcome, for example whether a customer canceled a subscription or this transaction is fraudulent, or it can be predicting a quantity of outcome for example dollars spent, hours played. Unsupervised learning models are different since they don't have a target variable. A good example where they are used - is to do clustering where the observations are grouped to similar groups. This is broadly applied in customer segmentation. We will dive into this topic in the next lesson.

3. Supervised learning types

As discussed, there are two types of supervised learning models - classification, when the target variable is categorical, or regression, when the target variable is continuous. Here are some examples of classification: We can predict if a customer will cancel their service subscription, or if a transaction is fraudulent. In a situation where we have more than 2 classes, we could trying to predict a profession of a user based on their browsing and search history. Regression examples are continuous variables like the number of product purchases, gaming hours or dollars spent.

4. Data collection

Machine learning teams collect data from the systems, customers and other sources to get the best prediction of the outcome variable. For example, in the case of customer purchase predictions we would collect customer information, purchase history, browsing history, device details, usage frequency and other information to use in our predictive models.

5. Classification example

Now, let's begin with a classification example. Here, we have a sample historical dataset with a list of transactions and some data collected for each, plus a target variable defining if the transaction was fraudulent or not.

6. Classification - training

We will use the input features here,

7. Classification - learning

to learn rules to predict the fraud probability on unseen data. This means the machine learning models learn rules on data that is labeled. That is, for these specific transactions, we know for sure if they were fraudulent or not. We then use the rules learned to predict if new transactions are likely to be fraudulent or not.

8. Classification - unseen data

Here, we see new data without labels. Once the model is trained, we will apply it on these new transactions, and based on their information,

9. Classification - prediction

predict fraud probability for each.

10. Regression example

Great, let's look into another example, this time it's regression. Here we have customer level data - similar to the one we used in the causal model example. We will try to predict next month's spending amount.

11. Regression - training

Just as with classification, we use the input features

12. Regression - learning

to learn rules to predict the next month's spending amount on unseen data. This means the machine learning models learn rules on data that is labeled, that is, we know for sure what the spending amount of this customer was. Then we use those rules to predict what customers are likely to spend next month.

13. Regression - unseen data

Here, we have new recent data, and use it

14. Regression - prediction

to predict next month's spending amount.

15. Regression - actual prediction

One thing I didn't tell you - just as in the causal model case, this is an actual prediction based on a sample dataset I've created.

16. Let's practice!

Great progress! Let's now dive into some exercises to test your knowledge!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.