Lesson 2 — Big Data Diagnostics and Analytics Horizons
The six V's are problem clues
Do not learn the V's as a recital. Ask what management pressure each one creates.
| V | Deciding question | Typical management implication |
|---|---|---|
| Volume | How much data? | storage, loading, scanning, computation must scale |
| Velocity | How fast is it generated/captured/needed? | latency and ingestion architecture matter |
| Variety / Variability | How many forms/structures? | relational-only assumptions may fail |
| Viscosity | How hard is it to use or integrate? | mapping, identifiers, timing, semantics can dominate effort |
| Volatility | How fast does it change/how long useful? | refresh, retention, model timing must fit the use |
| Veracity | How trustworthy is it? | source evaluation, profiling, DQ become prerequisites |
Examples: - “billions of events” → Volume. - “must react in seconds” → Velocity. - “tables + logs + text + images” → Variety. - “different IDs and meanings make sources hard to combine” → Viscosity. - “feed loses value after one hour” → Volatility. - “provider cannot explain sampling or provenance” → Veracity.
Descriptive → Predictive → Prescriptive
- Descriptive: What happened? Why did it happen? Think historical BI/rear-view mirror.
- Predictive: What is likely to happen? Output may be a probability, forecast, or risk score.
- Prescriptive: What should we do to influence the outcome? Output is a recommended action.
Changed fact: “Customer has 72% churn probability” is predictive. Add “send Offer B because it is most likely to prevent churn,” and the answer becomes prescriptive.
Operational analytics is a use context
Operational analytics applies models/analytics to current operational streams and can trigger alerts/actions. A live machine-failure score is predictive in analytic intent; continuously applying it to sensor streams and automatically shutting down equipment is operational analytics.
Source: pp. 474–483.