Skip to content

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

Orchid Logistics built a model to predict late deliveries. It achieves 99% accuracy on the training data but only 62% on new data it has not seen. What is the most likely problem, and the best remedy?

The model is overfitted: it has learned noise in the training data, so it performs well there but poorly on unseen data. The remedy is to validate on separate test data and simplify the model or add data. Underfitting would show poor results on both sets.

  1. AUnderfitting; add more irrelevant variables
  2. BOverfitting; validate on separate test data and simplify the model or use more dataCorrect
  3. CSampling error in the test data only; ignore the result
  4. DBias from using supervised learning; switch to clustering

Explanation

A big gap between training and unseen-data performance indicates overfitting, where the model has learned noise specific to the training set. Remedies include holding out test data, cross-validation, simplifying the model and adding data. Underfitting would give poor results on both sets.

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