Definition
Bias is a systematic deviation that distorts the results of an analysis, model or decision-making process.
Human Explanation
A model or analysis may appear technically correct while producing systematically misleading or unfair results because the data or the way it is used does not adequately represent the situation.
Why it Matters
Bias can lead to incorrect decisions, unequal outcomes, discrimination and loss of confidence in analytical models.
Conceptual Boundary
Bias does not arise only from an algorithm. It may already exist in what is measured, who is represented, how data is collected, how a target is defined or how results are interpreted. Not every difference between groups is proof of bias, but unexplained systematic differences require examination.
Practical Perspective
Bias should be assessed during problem definition, data collection, model design and validation, and then monitored after deployment. A model can be accurate on average while performing poorly for an important subgroup.