Mislabeled Data in Datasets CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11application · medium · select all that apply

   A team is dealing with a dataset that has a moderate amount of mislabeled data. They want to handle the mislabels to improve model performance. Which of the following are valid handling strategies? (Select all that apply.)

   • ARemove the mislabeled instances
   • BIgnore the mislabels because they are few
   • CIncrease the model's complexity to memorize the labels
   • DRelabel the mislabeled instances
   • EUse robust learning techniques
   Check answers Discussion Select an answer first
2. 12expert · hard

   A data scientist is using a model-based approach to detect mislabeled data in a large dataset. They train a model and compute the loss for each training instance. They plan to flag instances with the highest loss. Which of the following is a potential pitfall of this approach?

   • AHigh-loss instances may be hard examples, not mislabeled
   • BThe approach requires a separate validation set
   • CThe approach only works for binary classification problems
   • DThe model will overfit the mislabeled instances and have low loss on them
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   Which of the following is a common source of mislabeled data in ML datasets?

   • AHuman annotators making mistakes when labeling data
   • BUsing a consistent labeling guideline across all annotators
   • CApplying feature scaling before model training
   • DSplitting data into training and test sets randomly
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   An e-commerce company is building a product recommendation model. They discover that 5% of the product categories in the training data are mislabeled. The cost of relabeling is high, and the company wants to minimize the impact on model performance without discarding too much data. Which strategy is most appropriate?

   • ARelabel only the instances that the model is most uncertain about
   • BUse a robust learning technique that is tolerant to label noise
   • CIgnore the mislabels because 5% is too small to affect model performance
   • DRemove all instances with suspected mislabels from the dataset
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A medical imaging company is developing a model to detect tumors. The dataset has a 2% mislabel rate, but the mislabels are concentrated in a specific demographic group. The company must maintain high sensitivity for tumor detection and is subject to regulatory scrutiny. Which handling strategy is most appropriate?

   • AIgnore the mislabels because 2% is below the acceptable threshold
   • BUse a robust learning technique to mitigate the noise
   • CRemove all instances from the affected demographic group
   • DRelabel the affected instances using a panel of expert radiologists
   Reveal Solution Discussion Select an answer first