Skip to content

ACCA Strategic Professional · Strategic Business Leader · Machine learning, AI and robotics

Orlan Bank's credit-scoring model was trained on ten years of lending decisions, during which loans to applicants from certain postcodes were rarely approved. The model now declines most applicants from those postcodes, though they have strong finances. Which risk has most directly materialised?

The risk is algorithmic bias inherited from biased historical training data. The model learned from past lending decisions that disadvantaged certain postcodes and so reproduces that pattern, even for financially strong applicants. This is a data-quality and fairness problem, not drift or overfitting.

  1. AOverfitting to random noise caused by too few input variables
  2. BAlgorithmic bias inherited from biased historical training dataCorrect
  3. CModel drift caused by a sudden change in market conditions
  4. DLoss of the black-box property, making the model fully transparent

Explanation

The model has learned patterns embedded in past human decisions, so it reproduces that discrimination. This is bias in the training data carried into the algorithm. Drift concerns changes over time after deployment, not a flaw present in the original data.

Did you get it right without looking?

One question tells you little. A timed set on Machine learning, AI and robotics shows your real accuracy, how long you take and where you lose marks.

Practise Machine learning, AI and roboticsFree diagnostic mock

More Machine learning, AI and robotics questions