When to use machine learning
Machine learning is not magic and it may be not be beneficial to all data-related problems. It is important at the end of this introduction to clarify when machine-learning techniques are extremely useful:
- It is not possible to code the rules: a series of human tasks (to determine if an e-mail is spam or not, for example) cannot be solved effectively using simple rules methods. In fact, multiple factors can affect the solution and if rules depend on a large number of factors it becomes hard for humans to manually implement these rules.
- A solution is not scalable: whenever it is time consuming to manually take decisions on certain data, the machine-learning techniques can scale adequately. For example, a machine-learning algorithm can efficiently go through millions of e-mails and determine if they are spam or not.
However, if it is possible to find a good target prediction, by simply using mathematical rules, computations, or predetermined schemas that can be implemented without needing any data-driven learning, these advanced machine-learning techniques are not necessary (and you should not use them).
