Data Science Fundamentals: Framing, Lifecycle, and Data Types
Domain scope
Data Science Fundamentals tests how you frame an analytics problem before touching data: the analytics lifecycle, big data characteristics, data types, and the roles that turn a business question into a testable model. Expect questions on phase ordering, deliverables, and what belongs in discovery versus model planning.
Core rules and concepts
- The Dell/EMC analytics lifecycle has six phases: Discovery, Data Preparation, Model Planning, Model Building, Communicate Results, Operationalize. Discovery ends with a problem statement, success criteria, and initial hypotheses-not a final model.
- Big data is often described by volume, velocity, variety, veracity, and value. Variety (structured, semi-structured, unstructured) drives storage and parsing choices more than volume alone.
- Distinguish descriptive, diagnostic, predictive, and prescriptive analytics. A dashboard is descriptive; a churn probability is predictive; an uplift-optimized retention offer is prescriptive.
- Unit of analysis must be fixed early: customer, account, transaction, device, or session. Mixing units causes double counting and leakage.
- Data types matter: nominal, ordinal, interval, ratio; discrete vs continuous. They determine valid summaries and model encodings.
Worked example
A telecom team asks, “Why are customers leaving?” Reframe: unit = subscriber-month; target = voluntary churn in the next 30 days; population = active postpaid subscribers; label window = 30 days after observation; features = usage and billing history before the observation date. Discovery output: hypotheses (price shock, network complaints), success metric (lift at top 10%), constraints (no future billing data), and data sources. Model planning then selects logistic regression and a validation split by time. The reasoning is that a clear unit, label window, and cutoff prevent target leakage and make the result actionable.
Traps
- Treating lifecycle phases as strictly linear; discovery and preparation often iterate.
- Confusing correlation with causation when a business asks “why.”
- Using future information to define the label or features.
- Reporting accuracy without a baseline or business metric.
- Calling any large dataset “big data” without checking variety and velocity.