Machine Learning: 2 Books in 1: Machine Learning for Beginners, Machine Learning Mathematics. An Introduction Guide to Understand Data Science Through the Business Application

Unsupervised Learning
Unsupervised machine learning uses unlabeled data. Data scientists don’t
know the output of yet. The algorithm must discover patterns on its own,
where patterns would otherwise be unknown. Find a structure in a place
where the structure is otherwise unobservable. The algorithm finds data
segments on its own. The model looks for patterns and structure in an
otherwise unlabeled and unrecognizable mass of data. Unsupervised
learning allows us to find patterns that would be unobservable without

---

computer scientists. Sometimes massive collections of data have patterns,
and it would be impossible to sift through all of it trying to find trends.
This is good for examining the purchasing habits of consumers so that you
can group customers into categories based on patterns in their behavior. The
model may discover that there are similarities in buying patterns between
different subsets of a market, but if you didn’t have your model to sift
through these massive amounts of complicated data, you will never even
realize the nature of these patterns. The beauty of unsupervised learning is
the possibility of discovering patterns or characteristics in massive sets of
data that you would not be able to identify without the help of your model.
A good example of unsupervised learning is fraud detection. Fraud can be a
major problem for financial companies, and with large amounts of daily
users, it can be difficult for companies to identify fraud without the help of
machine learning tools. Models can learn how to spot fraud as the tactics
change with technology. If you want to deal with new, unknown fraud
techniques, then you will need to employ a model that can detect fraud
under unique circumstances.
In the case of detecting fraud, it's better to have more data. Fraud detection
services must use a range of machine learning models to be able to combat
fraud effectively. Using both supervised and unsupervised models. It's
estimated that there will be about $32 billion in fraudulent credit card
activity next year, in 2020 Models for fraud detection classify the output
(credit card transactions) as legitimate or fraudulent.
They can classify based on a feature like time of day or location of the
purchase. If a merchant usually makes sales around $20, and suddenly has a

---

sale for $8000 from a strange location, then the model will most likely
classify this transaction as fraudulent.
The challenge of using machine learning for fraud detection is the fact that
most transactions are not fraudulent. If there was even a significant amount
of fraudulent transactions among non-fraudulent, then credit cards would
not be a viable industry. The percentage of fraudulent card transactions is so
small that it can create models that are skewed that way. The $8000
purchase from a strange location is suspicious, but it is more likely to be the
result of a traveling cardholder than fraudulent activity. Unsupervised
learning makes it easier to identify suspicious buying patterns like strange
shipping locations and random jumps in user reviews.