Lobbying Prediction Model

A machine learning project that predicts lobbying success using Canadian lobbying registrations data. The model determines whether a lobbying attempt will receive government funding based on features such as subject matter, registrant location, organizational structure, and target institution.

Results

Model Performance (Random Forest Classifier)

Metric	Score
Accuracy	71.5%
Precision	70.3%
Recall	71.7%
F1 Score	70.5%

Key Findings

When predicting government funding success (Class = 1), the model identified these top predictive features:

• Subject Matter: Budget (8.0%), Economic Development (7.8%), Infrastructure (5.9%)
• Target Institution: Natural Resources Canada (44.8%), Finance Canada (25.1%), Agriculture Canada (11.1%)
• Region (Area Code): Ottawa-613 (45.8%), Toronto-416 (24.5%)

Visualizations

Confusion Matrix

ROC Curve

Model Performance Charts

Features

• Data Preprocessing: Handling missing values with imputation, encoding categorical features, feature scaling with StandardScaler
• Class Imbalance Handling: RandomUnderSampler to address 33/67 class distribution
• Multiple Models: Logistic Regression, Random Forest, Gradient Boosting, XGBoost
• Evaluation Metrics: Accuracy, Precision, Recall, F1-Score, Confusion Matrix, ROC Curve, Lift Chart, Cumulative Gains

Project Structure

LobbyingPredictionModel/
├── lobbyingModel.ipynb    # Main Jupyter notebook with analysis
├── canadacities.csv       # Canadian cities reference data
├── images/                # Visualization outputs
│   ├── confusion_matrix.png
│   ├── roc_curve.png
│   └── cumulative_gains_lift.png
├── requirements.txt       # Python dependencies
├── LICENSE               # MIT License
└── README.md             # This file

Data

Primary Dataset (Not Included)

The primary dataset merged lobby.csv (~995K rows) is not included due to file size.

To obtain the data:

1. Visit the Office of the Commissioner of Lobbying of Canada
2. Download the lobbying registration exports
3. Merge the Primary Export and Subject Matters Export files
4. Save as merged lobby.csv in the project root

Data Dictionary

Column	Description
GOVT_FUND_IND_FIN_GOUV	Target variable: Government funding indicator (Y/N)
SUBJ_MATTER_OBJET	Subject matter of lobbying activity
RGSTRNT_ADDRESS_ADRESSE_DCLRNT	Registrant address (processed to city)
RGSTRNT_TEL_DCLRNT	Registrant phone (area code used as region proxy)
PARENT_IND_SOC_MERE	Parent company indicator
COALITION_IND	Coalition indicator
SUBSIDIARY_IND_FILIALE	Subsidiary indicator
DIRECT_INT_IND_INT_DIRECT	Direct interest indicator
INSTITUTION	Target government institution

Installation

1. Clone the repository:

git clone https://github.com/davidyang02/LobbyingPredictionModel.git
cd LobbyingPredictionModel

2. Install dependencies:

pip install -r requirements.txt

3. Obtain the primary dataset (see Data section above)

4. Run the Jupyter notebook:

jupyter notebook lobbyingModel.ipynb

Models Used

Model	Description
Logistic Regression	Baseline linear model
Random Forest Classifier	Ensemble learning with decision trees (best performer)
Gradient Boosting Classifier	Boosting-based decision tree model
XGBoost Classifier	Optimized gradient boosting

License

This project is open-source and available under the MIT License.