What is the term for bias introduced when the training dataset is not representative of the target population?

The term for bias introduced when the training dataset is not representative of the target population is selection bias. This occurs when the sample of data used for training does not accurately reflect the characteristics of the broader population that the model is intended to make predictions about. If certain groups are overrepresented or underrepresented in the training dataset, the model may learn patterns that do not hold true for the entire population, leading to inaccurate predictions when applied to real-world scenarios.

In contrast, generalization refers to a model's ability to perform well on unseen data, which is a desired outcome rather than a type of bias. Underfitting occurs when a model is too simple to capture the underlying patterns of the training data, resulting in poor performance for both the training data and unseen data. Overfitting happens when a model learns the training data too well, capturing noise along with the underlying distribution, which also leads to poor generalization to new data. These terms represent different concepts in model training and evaluation, while selection bias specifically addresses the issue of representativeness in the training data.