Part A: Select the correct splitting method

Choose Random, Grouped or Time-based

  1. Predict next month’s server failures from daily monitoring history.
  2. Classify independent support messages into departments.
  3. Classify medical records where one patient has multiple visits.
  4. Predict policy renewal using five years of chronological history.
  5. Classify exam images where many images belong to one candidate session.

Part B: Create an ML.NET split

Using today’s program:

  1. Change the dataset to 200 rows.
  2. Use seed 100.
  3. Reserve 20% for testing.
  4. Split the remaining 80% into 60% training and 20% validation overall.
  5. Print the three row counts.
  6. Run the program twice and confirm that counts remain consistent.

Hint

Required validation fraction from remaining data-
20 ÷ 80 = 0.25

Therefore, take 25% of the remaining 80% for validation.

Part C: Detect leakage

Identify what is wrong

1. Calculate average premium using all records.
2. Fill missing premium values.
3. Split the data.
4. Train the model.

Rewrite the order correctly.

Part D: Apply it to Assessment OS

Design splits for an exam-risk model

Prediction unit:
Grouping key:
Time column:
Training period:
Validation period:
Test period:
Reason random splitting may be dangerous:
Information that must remain unavailable:

Consider whether several 30-second video clips from one exam session should ever be distributed across different subsets.

PreviousNext