CLOUD-DIGITAL-LEADER practice question 24 of 35

An organization is using machine learning to make predictions. One of their datasets mistakenly includes mislabeled data. How will the prediction be impacted?

Full set: 411 questions
Answer options
A: Increased risk of privacy leaks
B: Increased risk of inaccuracy Correct answer
C: Decreased model compatibility
D: Decreased model training time
Correct Answer:
B. Increased risk of inaccuracy
Explanation

Supervised machine-learning algorithms derive their parameters from the correspondence between each input and its label. If some of those labels are wrong, the learning algorithm is optimized toward conflicting targets, so the resulting hypothesis generalizes poorly to new, correctly-labeled examples. The primary, direct consequence is lower predictive quality (higher error / lower accuracy); other characteristics of the model—privacy exposure, framework compatibility, or training duration—are not deterministically affected by label noise.

Why Incorrect

A. Privacy risk depends on presence of sensitive attributes or re-identification vectors, not on whether labels are wrong.
C. Compatibility is governed by model format (e.g., SavedModel, ONNX) and serving stack; mislabels do not alter this.
D. Training time is dominated by data volume and model complexity; wrong labels neither shorten nor reliably lengthen the compute steps.

References

1. Google Cloud, “Best practices for data preparation – Vertex AI,” Section ‘Label quality’, ¶2: “Incorrectly labeled examples can significantly reduce model accuracy.” (cloud.google.com/vertex-ai/docs/general/ml-use-best-practices)

2. Google Cloud, “Data Labeling Service documentation,” Quality guidelines, ¶1: “Low-quality or mislabeled data leads to poor model performance.” (cloud.google.com/data-labeling/docs/overview)

3. Algan & Ulusoy, “Image classification with deep learning in the presence of noisy labels,” Pattern Recognition, 2021, p. 1, DOI:10.1016/j.patcog.2020.107329 – demonstrates accuracy degradation under label noise.

4. Stanford University, CS229 Lecture Notes, “Supervised Learning with Noisy Labels,” Fall 2022, p. 3: “Label noise directly decreases the achievable test accuracy.”

5. Frick et al., “Mislabeling in supervised learning: Its effects and how to deal with it,” MIT OpenCourseWare supplemental reading, Section 2, p. 5 – concludes that erroneous labels manifest primarily as prediction error.

Scroll to Top