Statistical Hypothesis Testing
For machine learning practicals
Made by @onderkova
0 0
8
27
1: You are training a Linear Regression model (y = a + b · y). You want to test if the feature “House_Size” has any impact on the target “Price”. What is the correct Null Hypothesis?
80s
H0: b=0
H0: b!=0
H0: b>0
H0: b<0
2: You have a baseline model with 85% accuracy. You want to prove your new model is statistically better. Which set of hypotheses is correct?
80s
H0: μ=0.85 vs. H1: μ!=0.85
H0: The model achieves consciousness.
H0: μ≤0.85 vs. H1: μ>0.85
H0: μ>0.85 vs. H1: μ≤0.85
3: You are A/B testing a new “Buy” button color at significance level α = 0.05. You get resulting p-value 0.03. How will you decide?
80s
Accept H0. The colors perform identically.
Reject H0. The color change made a significant difference.
Fail to reject H0. The difference is just noise.
No idea, just make the button invisible.
4: You are building a mission-critical AI for a nuclear reactor. You need to be extremely sure before you change any settings. Which α value do you choose?
60s
α=1.00
α=0.10
α=0.05
α=0.01
5: A firewall fails to identify a virus and lets it in. What type of error is this?
60s
Type I
Type II
6: Which of the following are examples of a Type I Error?
80s
Medical test saying you are sick when you are healthy.
Spam filter marking a spam as real email.
FaceID unlocking your phone for a stranger.
Your code running without errors on the first try.
7: What does a p-value of 0.03 mean?
60s
There are 3% of errors in the dataset.
The model is 3% accurate.
The result is 97% correct.
There is a 3% chance the result happened by random chance assuming the Null is true.
8: You are building an AI for autonomous braking. H0: no obstacle. False Negative: car does not stop while seeing an obstacle. False positive: The car brakes for a shadow. How do you adjust α to make the car safer?
80s
Increase α, Type I error increases, Type II error decreases
Increase α, Type I error decreases, Type II error increases
Decrease α, Type I error decreases, Type II error increases
Decrease α, Type I error increases, Type II error decreases
9: You get a p-value of 0.04. This means there is a 96% probability that the Alternative Hypothesis is true.
40s
True
False
10: Slide
120s
11: You trained a model to detect fake product reviews. You want to check if the categorical feature “Review Platform” (Amazon, eBay, Etsy) is independent of the model’s prediction (Fake, Genuine). Which statistical test do you use?
70s
Z-test
Pearson Correlation Coefficient
Student’s T-test
Chi-Square Test of Independence
12: Comparing the average salary of Data Scientists vs. Software Engineers. You survey 100 random people from each job. Which test is appropriate?
70s
Paired T-test
One-Sample Z-test
Independent Two-Sample T-test
A coding battle
13: You want to test if data augmentation improves model accuracy. You train a model on Dataset 1 (with aug) and Dataset 1 (without aug). You repeat this for all 20 datasets. Which test is appropriate?
70s
Paired T-test
The Turing Test
Independent Two-Sample T-test
Chi-Square Test
14: When do you use a T-test instead of a Z-test?
60s
When sample size is small.
When sample size is huge.
When data is categorical.
When you lost your Z-table.
15: You test 1,000 random features against a target with α=0.05. How many features will likely appear significant purely by chance?
40s
0
5
50
500
16: What is the Multiple Comparison Problem?
80s
The statistical requirement to double your sample size for every new variable you add.
When your dataset has too many duplicate rows.
When the features in your dataset are too similar to each other.
Running many tests increases the chance of getting a "significant" result just by luck.
17: You are doing a Grid Search with many tests. How do you fix the Multiple Comparison Problem?
60s