DSS110 DATA ANALYTICAL LIFE CYCLE FLASHCARDS

How do data science projects fundamentally differ from traditional Business Intelligence projects?

Data science projects are more exploratory in nature.

Why is it important to have a well-defined process in a data analytics project?

It ensures a comprehensive and repeatable method for conducting analysis.

What is a common mistake made at the start of a data science project?

Rushing into data collection and analysis before properly framing the business problem.

What may happen if project sponsors' objectives do not match the available data mid-stream?

The project may need to revert to the discovery phase or be cancelled.

How does documenting a process provide credibility to a data science team's findings?

It demonstrates rigor in the methodology used.

The Data Analytics Lifecycle consists of _____ distinct phases.

six

In the Data Analytics Lifecycle diagram, what do the circular arrows between phases represent?

Iterative movement between phases until sufficient information is gathered.

Role: Business User

A domain expert who consults on project context and benefits from the final results.

Which role is primarily responsible for the genesis and funding of an analytics project?

The Project Sponsor.

What is the primary responsibility of the Project Manager in an analytics team?

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

Role: Business Intelligence Analyst

An expert in data feeds and KPIs who creates dashboards and reports from a reporting perspective.

What is the primary duty of a Database Administrator (DBA) in an analytics project?

Provisioning and configuring the database environment to support the team's needs.

How does the role of a Data Engineer differ from a Database Administrator?

The Data Engineer executes data extractions and manipulations rather than just configuring the environment.

Which role provides subject matter expertise for analytical techniques and data modelling?

The Data Scientist.

What is the primary objective of Phase 1 (Discovery) in the analytics lifecycle?

To learn the business domain, assess resources, and frame the problem.

In Phase 1, why does the team look into relevant business history?

To learn from past similar projects attempted by the organisation.

What are the five categories of resources that a team must assess during the discovery phase?

Technology, tools, systems, data, and people.

Term: Framing the Problem

The process of stating the analytics problem to be solved and sharing it with stakeholders.

Why is it important to establish failure criteria during the discovery phase?

It helps the team recognise when to stop unproductive effort.

Who should be identified as a key stakeholder in an analytics project?

Anyone who will benefit from or be significantly impacted by the project.

During a sponsor interview, what is the purpose of asking about industry issues?

To identify external factors that may impact the analysis.

Term: Initial Hypotheses (IHs)

Ideas formed during discovery that the team tests with data in later phases.

What is the purpose of capturing aggregate data sources during the discovery phase?

To provide a high-level understanding and quick overview of the data.

What dictates the specific tools and infrastructure a team needs for a project?

The type and structure of the data.

Under what condition can a team move from Phase 1 (Discovery) to Phase 2?

When they have enough information to draft an analytics plan for peer review.

What are the three main steps involved in Phase 2 (Data Preparation)?

Exploring, preprocessing, and conditioning data.

Term: Analytic Sandbox

A workspace where teams can explore data without interfering with production databases.

Why do data science teams prefer an expansive approach to data collection in a sandbox?

More data allows for a mixture of purpose-driven analyses and experimental approaches.

How does the 'ELT' approach differ from the traditional 'ETL' approach?

ELT loads raw data into the datastore before any transformations occur.

Why is the 'ELT' approach particularly valuable for fraud detection analysis?

It prevents the inadvertent cleansing of outliers that might indicate fraudulent activity.

Data preparation typically accounts for at least _____% of a data science project's time.

$50\%$

What is a common consequence of jumping straight to modelling without proper data preparation?

Teams realise the data does not support the models and must return to Phase 2.

Term: Data Conditioning

The process of cleaning data, normalising datasets, and performing transformations.

Why should a Data Scientist be involved in the data conditioning step?

Decisions made during conditioning significantly affect subsequent analysis.

What does Ben Shneiderman's mantra for visual data analysis advise?

Overview first, zoom and filter, then details-on-demand.

How does 'skewness' in data impact visual analysis?

It shows if the majority of data is heavily shifted toward one end of a continuum.

What is the primary goal of Phase 3 (Model Planning)?

To determine methods, techniques, and workflow for model building.

In Phase 3, what informs the selection of variables and analytical methods?

The initial hypotheses developed in Phase 1.

Why should a team research model planning in different industry verticals?

To get ideas from how others have solved analogous problems.

In model selection, why is it better to capture 'essential predictors' rather than every possible variable?

It identifies the most influential variables and avoids unnecessary complexity.

Term: Model (in data science)

An abstraction from reality that emulates real-world behaviour with rules and conditions.

What should a team document while choosing and constructing preliminary models?

The modelling assumptions being made.

What is the primary objective of Phase 4 (Model Building)?

To develop datasets for training and testing and to build and execute models.

What is the difference between 'training data' and 'test data'?

Training data is for initial experiments, while test data validates the approach.

Which two phases of the lifecycle often overlap and require frequent iteration?

Model Planning (Phase 3) and Model Building (Phase 4).

Why is it vital to record the logic and results of a model during the building phase?

Small decisions made during construction are easily forgotten once the project ends.

What should a team do if a model's parameter values do not make sense to domain experts?

They should re-evaluate the model's validity and contextual accuracy.

Name three commercial tools commonly used for model building.

SAS Enterprise Miner, SPSS Modeler, and Matlab.

Name three free or open-source tools commonly used for model building.

R, Python, and SQL.

What is the primary goal of Phase 5 (Communicate Results)?

To articulate findings and outcomes to stakeholders based on Phase 1 criteria.

In Phase 5, how should a team view a project that disproved its hypotheses?

As a success of the data's rigour rather than a personal failure.

How many significant findings are recommended to be shared with stakeholders in Phase 5?

The three most significant ones.

Why is it important to quantify the business impact of results in Phase 5?

It helps demonstrate the value of the findings to project sponsors.

What is the focus of Phase 6 (Operationalise)?

Deploying the model in a production environment, often via a pilot project.

Why is a pilot project preferred over a full-scale rollout in Phase 6?

It allows the team to learn performance constraints on a small scale and manage risk.

Who typically joins the project team during the Operationalise phase?

Engineers responsible for the production environment.

In Phase 6, what should be designed to handle inputs that fall outside the model's training range?

Alerts for when the model is operating 'out-of-bounds'.

What should a presentation for project sponsors focus on?

High-level takeaways and clean visuals to aid decision-making.

What technical graphs are appropriate for a presentation aimed at fellow data scientists?

ROC curves, density plots, and histograms.

How should the level of detail in a presentation change for a more executive audience?

The presentation should be more succinct and focused on value.

What triggers the need to retrain a model in a production environment?

A degradation in model accuracy identified through ongoing monitoring.

What is the purpose of the 'hold-out' dataset in Phase 4?

Testing the model to ensure its predictive power is robust.

During Phase 1, what is the role of the 'Analytics Sponsor' interview?

To shape the project's scope, goals, and desired outcomes.

How does a well-defined process help new members join an analytics team?

It provides a common framework that allows methods to be easily repeated.

What is the 'ETLT' process?

A combination of Extract, Transform, Load, and then Transform to facilitate analysis.

In Phase 2, what should be done if the team cannot obtain enough data of sufficient quality?

The team may be unable to perform the subsequent steps in the lifecycle.

Term: Multicollinearity

A data modelling challenge where input variables are strongly related to each other.

What is the specific focus of 'Model Selection' within Phase 3?

Choosing an analytical technique based on the end goal of the project.

What determines if a situation warrants a single model or a larger analytic workflow?

The structure of the dataset and the business objectives.

How do Phase 3 and Phase 4 compare in duration to Phase 2?

Phases 3 and 4 tend to move more quickly than the labor-intensive Phase 2.

What does a Data Scientist ensure regarding analytical methods in Phase 4?

That the methods are valid and fit for the business problem.

What must marketing teams understand regarding a customer churn model?

How to use the model's predictions to plan their interventions.

What are the four main deliverables typically produced at the end of a project?

Sponsor presentation, analyst presentation, code, and technical specifications.

How does the team determine if project results are a 'success' in Phase 5?

By comparing outcomes to the criteria developed during Phase 1 Discovery.

What is the risk of failing to translate technical analysis into business language?

Stakeholders will not see the value, and the project's effort will be wasted.