Mislabeled Data in Datasets CT-AI Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which of the following is a strategy for handling mislabeled data?

   • AIgnoring the mislabeled instances and training as usual
   • BIncreasing the learning rate to compensate for errors
   • CRelabeling the mislabeled instances with the correct labels
   • DAdding more features to the dataset
   Reveal Solution Discussion Select an answer first
2. 2expert · hard

   A team is training a model for fraud detection. The dataset has a small number of mislabeled instances, but they are highly influential because they are outliers. The team wants to correct the labels without losing the data. Which strategy is most appropriate?

   • ARelabel the mislabeled instances based on expert knowledge
   • BIgnore the mislabels because they are few in number
   • CUse a robust learning technique to down-weight the mislabeled instances
   • DRemove the mislabeled instances from the dataset
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A healthcare startup is building an image classifier to detect skin lesions from dermatoscope photos. The dataset was labeled by a team of junior interns using a custom annotation tool. The model's performance is lower than expected. Which source of mislabeled data is most likely to be the primary contributor?

   • AData entry mistakes when transferring labels from paper forms
   • BSensor noise in the dermatoscope causing image distortion
   • CAutomated labeling from a pre-trained model that was not fine-tuned
   • DHuman error during manual annotation due to insufficient expertise
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A data scientist is training a binary classifier for email spam detection. After training, they notice that the model misclassifies many emails that contain the word 'urgent'. Upon inspection, they suspect that some emails labeled as 'spam' are actually legitimate. Which detection technique would be most efficient to identify these specific mislabeled instances?

   • AManually review all emails containing the word 'urgent'
   • BRemove all emails containing the word 'urgent' from the dataset
   • CPerform statistical outlier analysis on the word frequencies of all emails
   • DTrain a separate model to predict the word 'urgent' and compare predictions
   Reveal Solution Discussion Select an answer first
6. 5application · easy

   A data engineer is reviewing a dataset of customer reviews labeled as either 'positive' or 'negative'. They notice that some reviews contain sarcasm and are labeled as 'positive' when the sentiment is actually negative. How should these instances be classified in terms of data quality?

   • AAs outlier data, because sarcastic reviews are rare
   • BAs missing data, because the sentiment is ambiguous
   • CAs mislabeled data, because the assigned label does not match the true sentiment
   • DAs duplicate data, because sarcasm can be interpreted in multiple ways
   Reveal Solution Discussion Select an answer first