An algorithm may be biased if not properly derived.
Factors that can lead to bias in machine learning:
(1) important data is not included in the analysis
(2) the data is from a subset of patients who are not representative of the general population
(3) the data from one or more minorities is not included
Problems can arise:
(1) if the algorithm is used for a population different from the population in which it was derived
(2) if it is used over time without adjusting for changes in the population
(3) if quality control is not performed
(4) if perceived errors are not investigated
where:
• If a minority was not included in the analysis, then bias would occur if the algorithm was used on a person from the minority.
