Guides: Data &amp; Digital Scholarship Tutorials: Knime

Procedures

How will Texas counties vote during the 2020 Presidential election?

Description

This hands-on workshop will guide participants through the process of constructing a decision tree to model 2016 votes for Donald Trump in Texas using various demographic attributes.

Using 2016 data, we will train our model and then test our model to measure its accuracy.

Using 2020 forecast data, we will use our model to predict 2020 votes for Donald Trump.

Learning Outcomes

Participants will be able to solve basic classification problems by constructing a basic decision tree model.

Participants will understand the basics of decision tree modeling.

Participants will be able to construct a workflow using the Knime Analytics Platform.

The Baylor Libraries’ Data Scholar Program provides a series of hands-on workshops designed to help Baylor researchers learn about

advanced data research methods, tools, and sources.

Data can be Text, Numbers, and Multimedia

Assessment/Feedback: https://baylor.qualtrics.com/jfe/form/SV_aghfDPWozuY9dJP

Knime stands for Konstanz Information Miner.

An open source data analytics package available for Windows, Mac, and Linux.

Drag & Drop graphical interface to assemble nodes for data processing, analysis, modeling, and visualization.

Download at https://www.knime.com/

Prepared Data

Variables included in the workshop dataset. All variables are on the voter tabulation district level for Texas.

% Votes for Donald Trump in the 2016 presidential election
Population per square mile (2016, 2020)
% White population (2016, 2020)
% African American population (2016, 2020)
% Hispanic population (2016, 2020)
% Adults with bachelor's degree (2016, 2020)
Median household income (2016, 2020)
County

2016 election data collected from the Texas Legislative Council Public FTP Server. 2016 demographic data collected from the American Community Survey (2012-2016) via IPUMS NHGIS. 2020 demographic forecasts collected from Simply Analytics (Baylor only).

QGIS Desktop was used to aggregate the block-group ACS data to the VTD level.

Step #1: Launch the Knime application
Step #2: Explore Interface Take a moment to explore the following components of the Knime interface: Knime Explorer Node Repository Node Description Workflow Outline Console
Step #3: Create a new workflow window Right-click LOCAL (Local Workspace) and select New KNIME Workflow
Step #4: Name the workflow Name: Decision Tree Trump Votes Click OK

Step #1: Import Excel data

In the Node Repository, expand IO and expand Read.

Drag Excel Reader (XLS) to Workflow Window.

Step #2: Configure Excel Reader Node

Double-click Excel Reader Node (or right-click/Configure)

Select file to read: Browse for Excel file
Select the sheet to read: <first sheet with data>
Table contains column names: Check
Row IDs: Table contains row IDs in column A
Under Preview: Click Reload

Traffic light should turn from red to yellow.

Step #1: Add bin node

Search Node Repository for bin

Drag Auto-Binner node to Workflow.

Step #2: Connect nodes

Drag from the tip of the Excel Reader node output to the edge of the Auto-Binner node input.

Step #3: Clean workflow

On the top toolbar, click the Open setting dialog button.

Under Node Connections, check Curved connections.
OK

On the top toolbar, click the Auto Layout button.

Step #4: Configure Bin node

Double-click Auto-Binner node.

All attributes in the Exclude left window except for Trump. Trump remains in the Include right window.
Binning Method: Fixed number of bins: 3
Bin Naming: Borders
Replace target column(s): Check

Step #1: Add partition node

Search Node Repository for partition

Drag Partitioning node to Workflow.

Step #2: Connect Auto-Binner node to Partitioning node

Step #3: Configure partitioning node

Double-click Partitioning node

Choose size of first partition: Relative[%]
Relative[%]: 80

All traffic lights should be at yellow.

Step #1: Add Decision tree learner node

Search Node Repository for decision.

Drag Decision Tree Learner node to Workflow.

Step #2: Connect Topmost Partitioning node to Decision Tree Learner node

Step #3: Configure decision tree learner node

Double-click Decision Tree Learner node

Class column: Trump
Quality Measure: Gini Index
Pruning Method: MDL (minimum distance length)

Step #1: Add Decision tree view node Search Node Repository for decision. Drag Decision Tree View (JavaScript) node to Workflow.
Step #2: Connect Decision Tree Learner node blue output to Decision Tree View (JavaScript) node blue input
Step #3: Run model On the top toolbar, click the Execute all executable nodes button.
Step #4: View Decision tree Right-click the Decision Tree View (JavaScript) node and select Interactive View: Decision Tree View

Step #1: Add Decision predictor node

Search Node Repository for decision.

Drag Decision Tree Predictor node to Workflow.

Step #2: Connect Partitioning node bottom output to Decision Tree Predictor input node.

Also...

Connect Decision Tree Learner node blue output to Decision Tree Predictor node blue input

Step #1: Add Scorer node Search Node Repository for scorer. Drag Scorer node to Workflow.
Step #2: Connect Decision tree predictor node to Scorer input node.
Step #3: Configure scorer node Double-click the scorer node First column: Trump Second column: Prediction (Trump) OK
Step #4: Run Model On the top toolbar, click the Execute all executable nodes button.
Step #5: Measure accuracy Right-click the Scorer node and select View: Confusion Matrix

Question:	If we adjusted the number of classes (bins) from 3 to 5, would the accuracy of the model increase or decrease?
Task:	Make it so and see if you are correct!