Skip to content

ACCA Strategic Professional · Advanced Performance Management · Data science and analytics

Hollin Bank uses a machine learning model to score staff for bonus eligibility. Review shows the model gives systematically lower scores to part-time employees because historical bonus data favoured full-time staff. What is the most appropriate description of this problem and response?

This is algorithmic bias inherited from biased historical training data. The right response is to review the data, test the model's outputs for fairness across groups and keep human oversight. Faster processing or deleting part-time records would not remove the bias and could worsen it.

  1. AAlgorithmic bias inherited from training data; review the training data and test outputs for fairness before useCorrect
  2. BData latency; move the model to real-time processing
  3. COverfitting caused by too few variables; add more financial metrics only
  4. DData duplication; delete the part-time records from the dataset

Explanation

The model reproduces bias embedded in historical data, so the remedy is to examine and correct the training data and test outcomes for fairness, with human oversight. Real-time processing does not remove bias, and deleting part-time records would exclude those staff and worsen unfairness.

Did you get it right without looking?

One question tells you little. A timed set on Data science and analytics shows your real accuracy, how long you take and where you lose marks.

Practise Data science and analyticsFree diagnostic mock

More Data science and analytics questions