Nave Bayesian Classifiers DATA-SCIENCE-FOUNDATIONS Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A data scientist is building a spam filter. From historical data, 10% of emails are spam. Among spam emails, 80% contain the word 'free'; among non-spam emails, only 5% contain 'free'. A new email contains 'free'. What is the posterior probability that this email is spam?

   • A0.10
   • B0.64
   • C0.50
   • D0.80
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A data scientist is choosing between Naive Bayes and logistic regression for a binary classification task with 50 features and 10,000 training samples. The features are highly correlated. The primary requirement is fast training and interpretability, but the model must also achieve good predictive accuracy. Which choice is most appropriate?

   • AUse Naive Bayes because it is the only model that can handle 50 features.
   • BUse Naive Bayes because it trains quickly and is interpretable, but be aware that correlated features may hurt accuracy.
   • CUse logistic regression because it handles correlated features better and is still interpretable.
   • DUse a deep neural network because it can model complex correlations.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is a common limitation of Naive Bayes?

   • AIt requires a large amount of training data to estimate parameters
   • BIt cannot handle categorical features
   • CIt assumes features are conditionally independent, which is often unrealistic
   • DIt can only be used for binary classification
   Reveal Solution Discussion Select an answer first
5. 19expert · medium

   A team is using Naive Bayes to classify documents into two categories. They notice that when a document contains the phrase 'not good', the model often classifies it as positive because the word 'good' has a high positive weight. They also observe that the model's overall accuracy is acceptable but it fails on these negations. Which approach would most effectively address this issue while preserving the benefits of Naive Bayes?

   • AApply Laplace smoothing with a very large alpha value.
   • BUse n-gram features (e.g., bigrams) so that 'not good' is treated as a single feature.
   • CSwitch to a support vector machine with a polynomial kernel.
   • DRemove the word 'not' from the vocabulary.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A Multinomial Naive Bayes model is trained on a corpus of news articles. In the training set, the word 'election' appears 0 times in the 'sports' class and 10 times in the 'politics' class. The data scientist applies Laplace smoothing with alpha=1. What is the smoothed probability of 'election' given the 'sports' class, assuming the total word count in the 'sports' class is 1000?

   • A1/1002
   • B1/1000
   • CCannot be determined without the vocabulary size
   • D0
   Reveal Solution Discussion Select an answer first