F. Unit: 2.4: Model Planning Phase

📊 Phase 3: Model Planning

Goal: Decide which models and techniques to use to analyze the data, based on business goals, data structure, and prior hypotheses. This is the blueprint phase — you’re designing the analytics plan before actually building models.

🔁 How This Phase Connects to Earlier Work

• This phase links back to Phase 1 (Discovery) where the team formed initial hypotheses based on the business problem and data.

• Now, in Phase 3, the team chooses which models or methods to test those hypotheses.

• The output of this phase is the analytics strategy for the next step: building the model (Phase 4).

🧩 Key Activities in Model Planning

✅ 1. Assess Data Structure

• Check the type and structure of data (e.g., text vs. transaction logs).

• This determines which tools and which analytics techniques are appropriate.

  • Example: Textual data may require NLP tools.

  • Transactional data may work well with regression or clustering.

✅ 2. Validate Against Business Objectives

• Confirm that the analytical techniques you choose will help you test or validate the business hypotheses.

• Techniques should help determine whether the assumptions from earlier phases hold true or not.

✅ 3. Decide on One Model or a Workflow

• Choose if a single model is enough or if a chain of techniques is needed.

  • Example: You might combine classification with clustering in a workflow.

• Tools like Alpine Miner can build model workflows visually and connect to Big Data systems (like PostgreSQL).

🔍 Learn from Industry Practices

Teams are encouraged to research how similar problems are solved in other industries. This can:

• Inspire model selection

• Suggest best practices for similar data and goals

📊 Table: Models Used by Industry

This kind of benchmarking helps teams brainstorm which candidate models to test.

🔎 2.4.1 – Data Exploration & Variable Selection

🧼 How It's Different from Phase 2

• In Phase 2 (Data Prep), data exploration focused on data cleaning and quality.

• In Phase 3, it’s about understanding relationships between variables to decide:

  • Which variables to include

  • Which methods to use

🛠 Tools Used

• Visualizations (e.g., charts, heatmaps) help show variable relationships.

• This high-level view guides smart variable selection and prepares for model testing.

💡 Stakeholder Input vs. Data Science

• Stakeholders often have hunches or hypotheses.

  • Some are right, some are wrong — sometimes for the wrong reasons.

• Data scientists must:

  • Validate those assumptions objectively

  • Use unbiased data exploration to confirm or reject hypotheses

  • Choose predictors (input variables) that correlate well with outcomes

⚠ Key Modeling Challenges

• Multicollinearity: When predictors are too closely related to each other

• Serial correlation: When data points are overly influenced by previous values

To deal with these:

• Use variable transformations

• Consider removing or combining inputs

• Select modeling techniques that handle correlation better

🧠 If Using Regression Models:

• Identify:

  • Predictor variables (inputs)

  • Outcome variables (targets)

• Ensure predictors are tied to outcomes, not just correlated with each other.

• Be ready to handle:

  • Correlation-only models (e.g., black-box predictions)

  • Causal models if explanatory power is required (e.g., forecasting or stress-testing)

🧠 2.4.2 – Model Selection

🔧 What Is a “Model”?

• A model is an abstraction of reality: a simplified version of what’s happening in your data.

• In data science, models follow a set of rules and conditions to:

  • Classify items

  • Predict outcomes

  • Group (cluster) similar items

  • Find associations or relationships

🎯 How to Choose:

• Match the technique to the project’s end goal.

• Choose from common model categories:

  • Classification – Predict categories (e.g., churn or not)

  • Clustering – Group similar entities

  • Association Rules – Discover co-occurrence patterns

• Narrow it down to a shortlist of viable candidates.

🧵 Structured vs. Unstructured Data

• Use different tools depending on the data:

  • Structured: Rows and columns (e.g., sales data)

  • Unstructured: Text, images, etc.

• MapReduce and other distributed methods help with unstructured Big Data (see Chapter 10).

📝 Always Document:

• Any assumptions made in model design

• Why certain models or variables were chosen

🧪 Tool Example: Prototyping Models

Often, models are first created in:

• R

• SAS

• Matlab

These tools:

• Are powerful for machine learning and statistics

• May not scale to very large datasets

• Can be limited in Big Data applications

🧰 2.4.3 – Common Tools in Model Planning

🚀 Final Checklist Before Phase 4

To complete Phase 3 successfully, your team should:

• Select a modeling approach based on project goals

• Finalize which variables to include

• Create a diagram or written plan of the model workflow

• List tools to be used (e.g., R, SQL, SAS)

• Document any assumptions

• Have a clear direction for testing models in Phase 4

• candidate models: potential models for clustering, classifying, or finding relationships in data

• dataset structure: refers to the arrangement and organization of data used in the analysis process

• analytical techniques: methods and tools used to analyze and process data to achieve business objectives

• variable selection: the process of identifying essential predictors and variables to include in the model

• structured data: data organized in a specific format or schema, making it easier to analyze

• unstructured data: data that lacks a specific format or structure, often requiring additional processing before analysis