1. Machine learning principles
Now, we'll learn key principles of machine learning and its fundamental types.
2. Machine learning types
As we learned in the previous video, machine learning means applying statistical and computer science methods on data to achieve these three goals.
3. Supervised vs. unsupervised ML
There are two key types of machine learning. The first two are called "supervised machine learning", and the third - "unsupervised learning". We will now spend some time understanding them.
4. Supervised ML data structure
First, supervised machine learning. Here's a typical supervised modeling dataset. It's a customer transactions dataset. The goal is to predict if a transaction is fraudulent.
5. Target variable
In supervised learning there is a target variable which we want to predict - in our case this is the probability that this transaction is fraudulent. Supervised learning models are defined by the fact that they have a target variable which supervises what the model is optimizing for.
6. Input features
Then, for each transaction we have a number of columns with data points collected about that transaction. These are called input features.
7. Example input features
Here's a list of example columns on each transaction: count of past fraud transactions for this customer, time of transaction, number of declined transactions in the last 30 days for this customer, and the transaction amount. Finally, the actual fraud label is here also.
8. Using input features
Now, supervised machine learning models use the input features about the transactions,
9. Predicting target variable
to predict the target variable of interest, in our case - how likely it is that a transaction is fraudulent.
This is the fundamental principle of supervised machine learning models.
10. Unsupervised ML data structure
Next, unsupervised learning. Here we only have input features, but no target variable.
11. Unsupervised input features
So, unsupervised machine learning uses input features,
12. Unsupervised ML results
to identify groups of similar observations. For example, the model might segment transactions based on the money amount, currencies, payment device and other variables into different groups.
13. ML examples - Marketing
Let's look into some examples. We'll start with marketing.
Marketing teams are using machine learning to predict which customers are likely to purchase next month, and give them further incentives, or predict an expected customer lifetime value, so they could customize their targeting and treat potential high value customers with premium service.
Marketing uses unsupervised learning in many different ways.
The most popular one is customer segmentation.
14. ML examples - Finance
In finance, machine learning is applied to identify which transaction attributes are predictive of a potential fraud. Also, in banking, machine learning can be used to identify customer behaviors, for example if they will default on their mortgage in the next month.
Unsupervised learning is also used in finance to segment transactions and identify potentially profitable, risky and money losing types of transactions.
15. ML examples - Manufacturing
The manufacturing industry is utilizing machine learning and artificial intelligence incredibly fast. Robotics is the typical example, but this topic is outside the scope of our course.
Simpler examples are in quality control - predicting if certain items in production are faulty and should be manually inspected. Machine learning can also be used to read machine sensors and parameters like heat, electricity usage and so on, to identify which ones are likely to break and need maintenance.
Unsupervised learning is applied broadly in anomaly detection, for example, we can group the readings from equipment sensors and identify anomalies/outliers which were not assigned to a "typical" group and are likely to result in manufacturing malfunction.
16. ML examples - Transportation
Finally, transportation is full of machine learning examples. It's used to predict the expected delivery time of the parcel, or identify the fastest route for delivering the item, or even to predict demand each week, and appropriately prepare for demand spikes by stocking more items, renting more vehicles and hiring more workers to ensure demand can be met.
17. Let's practice!
That's a lot of information. Let's now go cement it by doing a couple of exercises!