Statistics and Data for AI Practice Test
Practise the statistics and data skills behind every AI project: averages and spread, correlation, A/B tests, missing data, leakage and evaluation metrics.
AI Skills
Before you start
- Questions
- 15
- Time
- 15 minutes
- Format
- Single section
- The timer starts when you press Start. When it reaches zero, your answers are submitted automatically.
- Use the numbered navigator to move between questions, and flag any you want to come back to.
- For typed answers, keep within the word limit. Answers over the limit are marked wrong, as in the real test.
- You can submit early. Your score, the correct answers and explanations appear straight away.
Statistics and Data for AI Practice Test: all 15 questions with answers
Prefer to work through the test on paper? Answer the questions, then open “Show answer” to check.
Choose the correct answer, A, B, C or D.
Q1.A dataset of salaries has a few extremely high values. Which average is least affected by them?
- A.Mean
- B.Median
- C.Maximum
- D.Sum
Show answer
Answer: B. Median
The median is the middle value, so a few extreme values barely move it.
Q2.What does the standard deviation measure?
- A.The most common value
- B.The total of all values
- C.How spread out the values are
- D.The number of rows
Show answer
Answer: C. How spread out the values are
A larger standard deviation means values are more spread out around the mean.
Q3.In a normal distribution, roughly what share of values lie within one standard deviation of the mean?
- A.About 50%
- B.About 95%
- C.About 99.7%
- D.About 68%
Show answer
Answer: D. About 68%
The 68–95–99.7 rule: about 68% within one, 95% within two and 99.7% within three standard deviations.
Q4.What does a correlation of −0.9 indicate?
- A.A strong negative linear relationship
- B.No relationship
- C.A weak positive relationship
- D.That one variable causes the other
Show answer
Answer: A. A strong negative linear relationship
Close to −1 means as one variable goes up, the other reliably goes down.
Q5.Ice cream sales and swimming accidents rise together. What is the most likely explanation?
- A.Ice cream causes accidents
- B.A third factor, hot weather, increases both
- C.Accidents cause ice cream sales
- D.The data must be wrong
Show answer
Answer: B. A third factor, hot weather, increases both
This is a confounder: correlation does not show causation.
Q6.What is the purpose of an A/B test?
- A.To test two databases
- B.To train two models on the same data
- C.To compare two versions by randomly assigning users to each
- D.To check spelling
Show answer
Answer: C. To compare two versions by randomly assigning users to each
Random assignment means differences in results can be attributed to the version, not to who saw it.
Q7.A result has a p-value of 0.01. What does this usually mean?
- A.There is a 99% chance the new version is better
- B.The test failed
- C.The sample was too small
- D.A difference this large would be unlikely if there were really no effect
Show answer
Answer: D. A difference this large would be unlikely if there were really no effect
The p-value is the chance of results at least this extreme if the null hypothesis were true.
Q8.Which of these is data leakage?
- A.Scaling features using statistics from the whole dataset before splitting into train and test
- B.Using a validation set
- C.Removing duplicate rows
- D.Plotting a histogram
Show answer
Answer: A. Scaling features using statistics from the whole dataset before splitting into train and test
Test-set information leaks into training, making evaluation look better than reality.
Q9.Where should the values used to fill in missing data come from?
- A.The test set
- B.The training set only
- C.Random numbers
- D.The final predictions
Show answer
Answer: B. The training set only
Calculating them from the training set avoids leaking test information.
Q10.Only 2% of transactions are fraud. Why is accuracy a poor metric here?
- A.Accuracy cannot be calculated
- B.A model that never predicts fraud would still be 98% accurate
- C.Accuracy is only for images
- D.Fraud data has no labels
Show answer
Answer: B. A model that never predicts fraud would still be 98% accurate
Use precision, recall or the precision-recall curve for imbalanced data.
Q11.What does min-max normalisation do?
- A.Removes outliers
- B.Sorts the data
- C.Rescales values to a fixed range, usually 0 to 1
- D.Converts text to numbers
Show answer
Answer: C. Rescales values to a fixed range, usually 0 to 1
Each value becomes (value − min) / (max − min).
Q12.What does a confusion matrix show for a classifier?
- A.The model's training time
- B.The network's layers
- C.The learning rate
- D.Counts of true positives, false positives, true negatives and false negatives
Show answer
Answer: D. Counts of true positives, false positives, true negatives and false negatives
From these counts you can calculate accuracy, precision and recall.
Q13.A fair coin is tossed twice. What is the probability of two heads?
- A.0.25
- B.0.5
- C.0.75
- D.1
Show answer
Answer: A. 0.25
Independent events multiply: 0.5 × 0.5 = 0.25.
Q14.Using the IQR rule, a value is often treated as an outlier if it is
- A.exactly equal to the median
- B.more than 1.5 × IQR below the first quartile or above the third quartile
- C.the most common value
- D.any negative number
Show answer
Answer: B. more than 1.5 × IQR below the first quartile or above the third quartile
The interquartile range (IQR) is Q3 − Q1; values beyond 1.5 × IQR from the quartiles are flagged.
Q15.Why is a sample used instead of a whole population?
- A.Samples are always more accurate
- B.Populations cannot be described
- C.Measuring everyone is often too costly or impossible, so a representative subset is used
- D.Samples need no randomness
Show answer
Answer: C. Measuring everyone is often too costly or impossible, so a representative subset is used
A well-chosen random sample lets you estimate population values at a fraction of the cost.