Final Assessment

AI for Classes 11–12 — Final Assessment

Application-heavy questions drawn from across all twelve modules, with a safety, ethics and privacy subset. Pass mark 75%. Attempt again after focused revision if needed.

1. How is this course's workload described transparently?

~12 hours guided learning plus ~8 hours practical and portfolio work.Different values in different parts of the app.Exactly 6 hours with no practical work.It is not stated at all.

2. How do AI, ML, deep learning and generative AI relate?

Deep learning is part of ML, part of AI; generative AI is a modern application; not all AI is ML.They are four unrelated fields.All AI is deep learning.Generative AI contains all of AI.

3. A teacher wants AI help with report comments and has a marksheet with names. What must they do first? (privacy)

Strip all identifiers (Student A/B/C) or use synthetic data.Paste the full marksheet with names.Add more personal details for accuracy.Upload student photos too.

4. In Python, why must `input(...)` often be converted with `int(...)` before arithmetic?

`input` returns text, and you cannot do arithmetic on text without converting it.Python cannot add numbers at all.`int()` makes the code run slower.It is never necessary.

5. How do you detect missing values in a Pandas DataFrame?

`df.isnull().sum()` counts missing values per column.`df.delete()` removes the DataFrame.Missing values cannot be detected.`print(df)` fixes them automatically.

6. What does the standard deviation of a dataset tell you?

How spread out the values are around the mean.The most frequent value.The number of rows.The correlation between two columns.

7. How can you tell overfitting from underfitting using train and test scores?

Overfitting has high train but low test accuracy; underfitting is poor on both.They look identical in the scores.Overfitting is poor on both.Only training accuracy matters.

8. Why is RMSE usually larger than MAE for the same predictions?

RMSE squares errors, amplifying large ones more than MAE does.RMSE ignores large errors.They are always equal.MAE squares the errors, not RMSE.

9. How are precision and recall calculated?

Precision = TP/(TP+FP); Recall = TP/(TP+FN).Precision = TP/(TP+FN); Recall = TP/(TP+FP).Both equal (TP+TN)/total.Neither can be computed from a confusion matrix.

10. How does the K-nearest neighbours algorithm classify a new point?

By the majority vote of its k nearest labelled neighbours.By picking a random class.By training a neural network.By always choosing the first class.

11. What are the two repeating steps of K-means clustering?

Assign points to the nearest centroid, then move centroids to the mean of their points.Split and delete the data repeatedly.Train a decision tree twice.It has no repeating steps.

12. What makes Orange Data Mining a "no-code" tool?

You connect visual widgets on a canvas instead of writing code.It writes all the code for you automatically and hides it.It cannot do machine learning.It only works online with an account.

13. When you load an image with OpenCV, what is it stored as?

An array of pixel numbers with a shape (height, width, channels).A single text label.A sound file.Nothing — images cannot be loaded in code.

14. What are the 5 Vs of big data?

Volume, Velocity, Variety, Veracity, Value.Volume, Voltage, Variety, Vision, Value.Video, Voice, Vector, Volume, Variety.Just Volume and Velocity.

15. What does an artificial neuron do to its inputs?

Multiplies inputs by weights, adds a bias, and applies an activation function.It thinks about the inputs like a brain cell.It stores the inputs without computing.It deletes the inputs.

16. How does a large language model decide what token to write next? (safety)

It predicts the most probable next token from context, so it can be confidently wrong.It looks up guaranteed facts in a database.It never makes mistakes.It copies a textbook exactly.

17. Does retrieval-grounded generation remove the need to verify AI outputs? (safety)

No — it reduces hallucinations but outputs still must be verified.Yes — grounded output is always true.Yes — so verification is never needed.Grounding makes outputs less accurate.

18. How should generative AI be taught according to the course's content rules? (ethics)

Vendor-neutrally, using products only as examples of the concepts.By teaching one commercial product as the whole technology.By promoting a specific paid tool.By presenting speculation about the future as fact.

19. Why use 5W1H and empathy mapping when scoping a capstone problem?

They keep the project authentic and human-centred and help scope it feasibly.They add paperwork with no benefit.They replace the need for any data.They guarantee a top mark automatically.

20. How should open-ended capstone project work be assessed? (ethics)

By a human reviewer against the rubric, not automated grading alone.By an AI grader with no human review.It should not be assessed.Only by counting the number of pages.