Food Policy Media Sentiment Analysis Dashboard

Model Performance Metrics

Dataset Summary

Time Series Overview

Study Overview

This dashboard presents results from a machine learning analysis of media sentiment and food policy.

Key Findings:

RandomForest ROC AUC: 0.72 ± 0.03
LightGBM ROC AUC: 0.70 ± 0.02
Target prevalence: 2.5%
~240 weekly observations

Data Sources:

GDELT media coverage
Census HPS food scarcity
Federal Register policies

Diagnostics

Output dir: /home/site/wwwroot/outputs_media_policy

Files detected: 13 / 13

Prediction Scores Over Time

Select Model:

Model Performance Curves

Curve Type:

ROC

Precision-Recall

Feature Importance Comparison

Number of features to display:

SHAP Value Comparison

Feature Statistics

Media Coverage Analysis

Media Coverage Controls

Select Topic:

Metric:

Volume

Sentiment

Both

Date Range:

About This Dashboard

Research Question

Can media sentiment and topical volume from publicly available sources be used to predict near-term changes in household food insecurity in the United States?

Methodology

Data Sources: GDELT media database, Census HPS, Federal Register
Models: Random Forest and LightGBM classifiers
Validation: 5-fold time series cross-validation
Target: Policy events within 14-day horizon

Key Findings

Media volume more predictive than sentiment alone
Food prices and SNAP coverage are strongest predictors
Both models achieve ~70% ROC AUC
Short-term lags (1-2 weeks) most informative

Author

Jasmine Motupalli
Daniels College of Business, University of Denver
FIN-6305: Applied Quantitative Methods