Part A: Identify the learning approach

Classify each requirement:

  1. Predict whether an insurance policy will renew.
  2. Estimate next month’s dairy production quantity.
  3. Group psychometric-test participants by response patterns without predefined profiles.
  4. Route a complaint to one of six departments.
  5. Learn which non-certification practice question should be shown next.
  6. Calculate a legally defined CSR obligation.
  7. Find groups of servers with similar CPU, memory and connection patterns.
  8. Predict the expected number of SQL connection failures tomorrow.

Choose from:

Deterministic C#/SQL
Binary classification
Multiclass classification
Regression
Clustering
Reinforcement learning

Part B: Define the label

For every supervised problem above, write:

Prediction unit:
Features:
Label:
Label data type:
Prediction moment:

Example:

Prediction unit: One expiring policy
Features: Claims, complaints, premium, renewal history
Label: DidNotRenew
Label data type: Boolean
Prediction moment: 30 days before expiry

Part C: Design a clustering problem

For APW Insurance, complete:

Business exploration goal:
One row represents:
Available features:
Number of initial clusters to test:
Who will interpret the clusters:
How the discovered groups might be used:
Sensitive features to exclude or review:

Remember: do not create target labels before clustering. The purpose is to discover candidate structure.

Part D: Validate an RL proposal

For an adaptive learning application, answer:

Agent:
Environment:
State:
Possible actions:
Immediate reward:
Long-term reward:
Unsafe action:
Can we train in a simulator?
Is human approval required?
Could supervised learning solve it more safely?

Part E: Run the C# selector

Create a console application and test these profiles:

 
var customerGrouping = new ProblemProfile(
    HasHistoricalLabel: false,
    LabelIsCategory: false,
    LabelIsNumericQuantity: false,
    NeedsHiddenGroups: true,
    HasSequentialActionsAndRewards: false);

var resolutionTime = new ProblemProfile(
    HasHistoricalLabel: true,
    LabelIsCategory: false,
    LabelIsNumericQuantity: true,
    NeedsHiddenGroups: false,
    HasSequentialActionsAndRewards: false);

Expected results:

UnsupervisedClustering
SupervisedRegression
 
PreviousNext