Which of the following is one of the reasons for data mislabelling?

Data mislabeling, or incorrect annotation, is a significant data quality issue that can severely degrade an AI model's performance. A primary cause of mislabeling is a lack of domain knowledge by the person or system performing the labeling. If the annotator does not understand the specific context, nuances, or subject matter of the data, they are highly likely to assign incorrect labels. For example, a medical image annotator without medical training may misclassify tissue types.

B. Expert knowledge is the opposite of a cause; it is a prerequisite for accurate labeling and helps prevent mislabeling.
C. Interoperability error relates to problems with different systems exchanging data, which is a system integration issue, not a direct cause of incorrect data annotation.
D. Small datasets are a data quantity issue. While mislabeling has a greater impact on small datasets, the size itself does not cause the labels to be incorrect.

1. ISTQB® Certified Tester AI Testing (CT-AI) Syllabus Version 1.0

Section 3.2.2

page 18: Discusses data quality issues

including incorrect labeling

and implies the need for subject matter expertise for correct labeling.

2. Northcutt

C.

Athalye

A.

& Mueller

J. (2021). Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks

1. Section 3.1 discusses sources of label errors

including ambiguity that requires domain expertise to resolve. (Sourced from MIT research). Link: https://dspace.mit.edu/handle/1721.1/146459