Prediction models (supervised learning)

Machine Learning for Business

Karolis Urbonas

Head of Machine Learning & Science, Amazon

Supervised vs. Unsupervised

Supervised models

Predicting class/type of an outcome (e.g. subscription cancellation, fraud, purchase) - CLASSIFICATION

Predicting quantity of an outcome (e.g. dollars spent, hours played) - REGRESSION

Unsupervised models

Clustering - grouping observations into similar groups or clusters (e.g. customer or market segmentation)

Supervised learning types

Classification - Target variable is categorical (discrete) (class of outcome) (classification)

Will the customer cancel a service subscription?

Is this transaction fraudulent?

What is the profession of this user?

Regression - Target variable is continuous (amount of outcome) (regression)

Number of product purchases next month

Number of gaming hours next year

Dollars spent on insurance

Data collection

Machine learning teams should collect all available data to predict desired outcome with the highest degree of accuracy e.g. in case of purchase predictions:

Customer information

Purchase history, cancellations, order amount

Browsing history, logs, errors

Device details and location

Product/service usage frequency

And others...

Classification example

classification - data

Classification - training

classification-training

Classification - learning

classification-learning

Classification - unseen data

classification-unseendata

Classification - prediction

classificaiton-prediction

Regression example

regression-data

Regression - training

regression-training

Regression - learning

regression-learning

Regression - unseen data

regression-testdata

Regression - prediction

regression-prediction

Regression - actual prediction

regression-actual

Let's practice!

Machine Learning for Business