A term used specifically in machine learning to refer to how closely a predicted value for some data matches the actual value. Typically, the ratio of correctly predicted observations to all the observations in the dataset is used as the accuracy metric. For instance, the accuracy of a binary classification model would be 90 per cent if it correctly predicted 90 out of 100 samples.
When assessing the privacy implications of a machine learning model, accuracy can be a factor to consider. 100 per cent accuracy might be undesirable in some circumstances as it could imply that the model can reliably infer private or sensitive information about people. This also means that accuracy information might itself be disclosive and might need to be subject to disclosure control.
p. 5Further reading
See also: INFERENCE
Article 29 Data Protection Working Party, 2014. Opinion 05/2014 on anonymisation techniques. Available from: https://ec.europa.eu/justice/article-29/documentation/opinion-recommendation/files/2014/wp216_en.pdf.
Yin, M., Wortman, V.J. and Wallach, H., 2019. Understanding the effect of accuracy on trust in machine learning models. In: Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–12. https://doi.org/10.1145/3290605.3300509.
-
Article 29 Data Protection Working Party, 2014. Opinion 05/2014 on anonymisation techniques. Available from: https://ec.europa.eu/justice/article-29/documentation/opinion-recommendation/files/2014/wp216_en.pdf.
-
Yin, M., Wortman, V.J. and Wallach, H., 2019. Understanding the effect of accuracy on trust in machine learning models. In: Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–12. https://doi.org/10.1145/3290605.3300509.
