DSS110 DATA SCIENCE ANALYTICAL LIFE CYCLE

Business User

Understands the domain area and usually benefits directly from the project results.

Project Sponsor

Responsible for the genesis of the project, providing high-level requirements, vision, and funding.

Project Manager

Ensures key milestones and objectives are met on time and at the expected quality.

Business Intelligence Analyst

Provides business domain expertise based on deep understanding of data, KPIs, and key metrics.

Database Administrator (DBA)

Provisions, configures, and secures the database environment to provide team access to required data.

Data Engineer

Tunes SQL queries, builds pipelines, and cleanses data to ensure it is structured and ready for analytical use.

Data Scientist

Applies statistical modeling and advanced analytics techniques to solve business problems and extract actionable insights.

Discovery (Phase 1) - Core Purpose

The phase where the team learns, investigates the business problem, develops domain context, identifies data sources, and formulates initial hypotheses.

Discovery - Core Question

"Do I have enough information to draft an analytic plan and share for peer review?"

Problem Framing

The process of clearly stating the analytics problem, defining main objectives, business deliverables, success criteria, and failure criteria.

Initial Hypotheses (IHs)

Formulating testable ideas early in the project that serve as the foundation for analytical tests and findings.

Analytic Plan

A comprehensive document created at the end of Phase 1 that tests the team's grasp of the business problem, data scope, and sets success criteria.

Data Preparation (Phase 2) - Core Purpose

The phase involving steps to explore, preprocess, clean, and condition data prior to modeling.

Data Preparation - Core Question

"Do I have enough good quality data to start building the model?"

Data Preparation - Time Investment

Consumes at least 50% of a data science project's total time due to its highly iterative nature.

Analytic Sandbox

A dedicated workspace separate from live production databases that allows data scientists to explore high volumes/varieties of data safely.

ELT vs. ETL (ETLT)

Advocates loading raw data before transforming (ELT/ETLT) to preserve raw data nuances and avoid filtering out key outliers (e.g., fraud indicators).

Dataset Inventory Matrix

A tool used to catalog datasets across 4 categories: Available & Accessible, Available but Not Accessible, Data to Collect, and Third-Party Data.

Data Conditioning

Preprocessing step involving cleaning, normalizing, joining, and transforming datasets, while handling missing values and data type alignment.

Ben Shneiderman's Mantra

"Overview first, zoom and filter, then details-on-demand" — a guideline for surveying and visualizing data.

Model Planning (Phase 3) - Core Purpose

The phase where the team explores data relationships, selects key variables, and determines candidate models and analytical techniques.

Model Planning - Core Question

"Do I have a good idea about the type of model to try? Can I refine the analytic plan?"

Data Exploration & Variable Selection

Evaluating variable relationships to select essential predictors while avoiding multicollinearity and serial correlation.

Model Building (Phase 4) - Core Purpose

The phase where datasets are split into training/test sets to execute, fit, evaluate, and refine analytical models.

Model Building - Core Question

"Is the model robust enough? Have we failed for sure?"

Training Dataset vs. Hold-out Test Dataset

The training dataset is used to build and train the model; the test (hold-out) dataset is used to validate and score model performance.

Model Refinement Iteration

The process of iterating back and forth between Phase 3 (Planning) and Phase 4 (Building) to tune inputs and optimize performance.

Commercial Model Building Tools

SAS Enterprise Miner, SPSS Modeler, Matlab, Alpine Miner, STATISTICA, Mathematica.

Open-Source Model Building Tools

R / PL/R, Python, Octave, WEKA, SQL.

Communicate Results (Phase 5) - Core Purpose

The phase where the team determines success/failure, compares outcomes with Phase 1 hypotheses, and articulates key findings to stakeholders.

Redefining "Failure" in Data Science

Failure is not a true negative outcome, but rather a failure of the data to adequately accept or reject a given hypothesis.

Key Best Practice in Phase 5

Record all findings and select the top 3 most significant ones to present to stakeholders along with business impact.

Operationalize (Phase 6) - Core Purpose

The phase where the team deploys the analytical model into a production environment, starting with a controlled pilot.

Pilot Deployment

Undertaking a small-scope rollout before enterprise-wide scaling to manage risk and assess runtime performance constraints in production.

Model Out-of-Bounds Monitoring

Setting automated alerts in production for when live input data drifts outside the training range, indicating the model needs retraining.

Deliverable: Presentation for Project Sponsors

High-level, succinct briefing with clean visuals focused on business value and strategic decision-making.

Deliverable: Presentation for Analysts

Detailed technical overview covering methodology, process changes, and diagnostic graphs (ROC curves, density plots, histograms).

Deliverable: Code & Technical Specs

Well-documented code scripts and architectural implementation specifications for production engineers.