Machine learning algorithms leverage data to mimic intelligent behaviour for specific tasks, varying from detecting anomalies in industrial processes to movie recommendations. Recent technological advances enabled efficient transfer, storage, and process of data. These advances also impacted machine learning algorithms by allowing an ever-growing increase in complexity, which sometimes means increased accuracy.
However, to effectively learn from fast data, it is also essential to account for changes that may have catastrophic effects on the machine learning algorithms. For example, an algorithm may signal legit credit card usage as a fraud because the users' behaviour has changed.
The reason behind these changes may not be accessible to the algorithm, making it unable to react to them immediately. For example, a different online credit card usage pattern may be because customers cannot leave their homes.
This paper combines theoretical and practical aspects of machine learning for data streams. We present examples using the methods implemented in the Massive Online Analysis (MOA) framework and river (Python) framework.

The learning outcomes for this paper are linked to Washington Accord graduate attributes WA1-WA11. Explanation of the graduate attributes can be found at: https://www.ieagreements.org/

https://mitpress.mit.edu/books/machine-learning-data-streams

There is an official online version available at https://moa.cms.waikato.ac.nz/book/