Dataset Bias

Dataset bias refers to systematic skew in what a dataset contains or how it was filtered, such that patterns learned or measured from it fail to represent the full population of interest. It commonly arises from geographically lopsided collection — for example a sketch dataset drawn disproportionately from the US — compounded by preprocessing steps like recognizer-based filters that quietly discard culturally atypical examples before any analysis begins.