Objective 2.1 Recognize Relative Impact of Data Quality and Size to Algorithms AIP-210 Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   Which of the following is a typical effect of outliers on a machine learning model?

   • AThey have no impact on model performance if the dataset is large enough.
   • BThey always improve model performance by providing more diverse examples.
   • CThey can skew the model's parameters, such as the mean and variance used in normalization.
   • DThey only affect models that use distance-based metrics, not tree-based models.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A data analyst is preparing a dataset for a binary classification task. They run a profiling script that outputs the following for each feature: count of non-null values, mean, standard deviation, min, max, and number of unique values. Which data quality issue can this summary directly reveal?

   • ALabel noise, by examining the standard deviation of the target
   • BMissing values, by comparing the count of non-null values to the total number of rows
   • CData leakage, by comparing min and max values between train and test sets
   • DDuplicate rows, by looking at the number of unique values
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A startup has two options for building a sentiment analysis model: Option 1 is a dataset of 1 million social media posts scraped automatically, but with many mislabeled examples and duplicate content. Option 2 is a dataset of 50,000 carefully curated and manually labeled posts. The team has limited compute and time. Which option is more likely to produce a reliable model?

   • AOption 1, because more data always leads to better model performance
   • BOption 1, because deep learning models require millions of examples to work at all
   • COption 2, because smaller datasets always train faster and are easier to work with
   • DOption 2, because high-quality labels are more important than raw volume when the data is noisy
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   In the context of training data volume, what does 'diminishing returns' mean for a machine learning model?

   • AThe model's performance becomes completely independent of data size after a certain point.
   • BAdding more data eventually causes the model to perform worse.
   • CEach additional training example provides less improvement in model performance than the previous one.
   • DThe model requires exponentially more data to achieve the same improvement.
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A team is training a deep learning model for object detection. They have 100,000 images, but the images are of varying quality: some are blurry, some have incorrect bounding boxes, and some are duplicates. They are considering three options: (A) train on all 100,000 images as-is, (B) clean the dataset to 60,000 high-quality images, or (C) train on the 100,000 images but apply aggressive data augmentation. Which option is most likely to produce the best model?

   • AOption A, because more data always helps deep learning models
   • BOption B, because removing low-quality images reduces noise and improves the signal-to-noise ratio
   • COption C, because augmentation can compensate for all data quality issues
   • DOption B, but only if the team also uses a larger model
   Reveal Solution Discussion Select an answer first