Data Quality and Its Effect on the ML Model CT-AI Practice Questions (Page 6)

Questions 26–30

1. 26expert · hard

   A ride-sharing company is building a model to predict trip duration. The dataset contains a 'distance' feature with some extreme values (e.g., 1,000 km in a city where trips rarely exceed 50 km). The team is considering three approaches: (A) remove the extreme values, (B) cap them at 50 km, or (C) keep them and use a robust model. Which approach is MOST appropriate, given that the extreme values are confirmed data-entry errors?

   • AImpute the extreme values with the mean distance.
   • BKeep the extreme values — robust models can handle outliers without issue.
   • CCap the extreme values at 50 km — it preserves the rows and reduces their impact.
   • DRemove the extreme values — they are confirmed errors and would skew the model.
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   How can poor data quality, such as biased sampling, affect an ML model?

   • AIt always increases model accuracy
   • BIt can introduce bias into the model's predictions
   • CIt only affects the model's inference speed
   • DIt has no effect on model performance
   Reveal Solution Discussion Select an answer first
4. 28expert · hard

   A hospital is building a model to predict patient mortality. The dataset has 5% missing values in 'blood_pressure'. The team is considering two approaches: (A) delete all rows with missing values, or (B) impute with the median. The model will be used in a clinical setting where false negatives are extremely costly. Which approach is MORE appropriate, and why?

   • AImpute with median — it is the safest choice because it does not remove any potentially life-saving patient data.
   • BInvestigate the pattern of missingness and use multiple imputation or model-based methods.
   • CDelete rows — it ensures the model trains only on complete, high-confidence data.
   • DImpute with median — it preserves all data and is simple to implement.
   Reveal Solution Discussion Select an answer first
5. 29expert · hard

   A marketing team is building a model to predict customer response to a campaign. The dataset contains a 'response' column where 10% of the labels are known to be wrong (e.g., a customer who did not respond is labeled as responded). The team is considering three options: (A) remove the mislabeled rows, (B) correct the labels manually, or (C) train the model with the noisy labels. Which approach is MOST appropriate, given that manual correction is expensive and the mislabels are random?

   • ATrain the model with the noisy labels, using techniques robust to label noise.
   • BCorrect the labels manually to achieve 100% accuracy.
   • CRemove the mislabeled rows to ensure a clean training set.
   • DImpute the noisy labels with the majority class.
   Reveal Solution Discussion Select an answer first
6. 30foundation · easy

   A dataset contains records where the same customer's age is listed as 25 in one table and 35 in another. Which data quality dimension is violated?

   • ATimeliness
   • BAccuracy
   • CConsistency
   • DCompleteness
   Reveal Solution Discussion Select an answer first