Exam Targets 09–16
9. Data-mining technique selection
Profiling understands structure; reduction simplifies; association finds co-occurrence; clustering groups similar cases; self-organizing maps organize/visualize complex similarity. Match technique to the question.
Source: pp. 481–482.
10. Predictive vs Operational analytics
Predictive estimates likely outcomes. Operational analytics applies analytics/models to live/current operational streams and may trigger response.
Trap: “fast” prediction alone does not prove operational analytics.
Source: pp. 481–483.
11. Source selection before ingestion
Evaluate origin, meaning, relationships, timing, grain, consistency, reliability, foundational value, privacy, bias, and feasibility before relying on a source.
Trap: cheapest/largest source automatically wins.
Source: pp. 485–487.
12. Privacy, filtering, and bias
Combining sources can re-identify; selection/filtering can exclude populations and bias results.
Clue: what becomes identifiable or systematically absent?
Source: pp. 486, 498–499.
13. Acquire/Ingest with Metadata
Capture origin, size, currency, content/structure, lineage, intended use, profiling/DQ context while onboarding.
Trap: add Metadata after modeling.
Source: pp. 486–487, 499.
14. DQ before integration
Assessment is an input to source choice, mapping, alignment, and model feasibility—not final cleanup.
Trap: “clean it later if the model fails.”
Source: pp. 487, 499–500.
15. Integrate/Align data for analysis
Align keys, semantics, timing, and granularity; Master/Reference Data may supply common context. A matching field name/key does not prove analytical alignment.
Source: pp. 487–488.
16. Hypothesis/model quality vs input-data quality
Good inputs cannot rescue bad assumptions; elegant models cannot rescue bad evidence. Diagnose whether failure is in reasoning/model or source evidence.
Source: pp. 487–489.