A security camera ML system is disproportionately flagging people from a particular ethnic group. Which kind of bias is causing this unfair outcome?

Sampling bias occurs when the data used to train the ML model does not accurately represent the real-world distribution of the population. In this case, the training data likely lacked sufficient examples of the particular ethnic group, leading the model to misclassify or disproportionately flag individuals from that group. Measurement bias relates to systematic errors in data collection, not the representativeness of the sample itself. Observer bias involves a researcher's expectations influencing observations, which is less relevant to an automated ML system's output. Confirmation bias is the tendency to interpret new evidence as confirmation of one's existing beliefs, a human cognitive bias, not a machine learning model's inherent flaw in this context.