What does Machine Learning mean?

Machine learning refers to methods that derive their rules from data instead of having them fixed in advance. From many examples with known outcomes, a model emerges that can later solve the same task on new data. The difference to classic programming is that the rule only takes shape during training, rather than being formulated by a person beforehand.

In supervised learning, the correct outcome is known for every example, such as the later result of a process. In unsupervised learning, these labels are missing, and the method searches on its own for groups or anomalies. In reinforcement learning, the rule emerges from feedback on actions taken. In all three cases, part of the data is held back and used only after training, to measure how well the model performs on cases it has not seen.

Machine learning pays off wherever a relationship exists but cannot be written down as a rule. Examples include forecasting demand and utilization, detecting anomalies in booking data, and classifying incoming text. Where a fixed regulation already determines the decision, a programmed rule remains the better choice.

The advantage over a manually maintained rule lies in the number of factors involved. A model processes dozens of features at once and weighs them based on the data. When conditions change, the model is retrained on new data instead of rewriting the rule.

A model only knows the cases that appeared in its training data. If groups are missing, or if the data carries an earlier pattern of unequal treatment, a bias emerges that continues unchecked once the model is in operation. The composition of the training data must therefore be documented.