Banking Marketing Campaign Analysis and Prediction

Project Overview

This project analyzes the effectiveness of direct marketing campaigns for a banking institution and predicts customer subscription to term deposits based on historical campaign data. The model leverages various machine learning techniques, feature engineering, and statistical analysis to derive insights and improve prediction accuracy.

Key Features

• Data Preprocessing: Handling missing values, outliers, and categorical encoding.
• Feature Engineering: Creating new features to enhance model performance.
• Exploratory Data Analysis (EDA): Detailed visualizations to uncover trends and patterns.
• Model Training: Comparison of Logistic Regression, Random Forest, and XGBoost models.
• Performance Metrics: Evaluating models on accuracy and F1-score.
• Submission Generation: Creating predictions in the required format for submission.

Dataset Description

The data is related to direct marketing campaigns of a banking institution. Campaigns were conducted via phone calls, often involving multiple contacts per client to determine whether they would subscribe to a term deposit.

Files

• train.csv: Training dataset
• test.csv: Test dataset
• sample_submission.csv: Sample submission file in the correct format

Input Variables

1. last contact date: Date of last contact
2. age: Age of the client (numeric)
3. job: Type of job (categorical)
4. marital: Marital status (categorical)
5. education: Level of education (categorical)
6. default: Has credit in default? (binary)
7. balance: Average yearly balance in euros (numeric)
8. housing: Has housing loan? (binary)
9. loan: Has personal loan? (binary)
10. contact: Contact communication type (categorical)
11. duration: Last contact duration in seconds (numeric)
12. campaign: Number of contacts performed during this campaign and for this client (numeric)
13. pdays: Days passed since the client was last contacted from a previous campaign (numeric)
14. previous: Number of contacts performed before this campaign (numeric)
15. poutcome: Outcome of the previous marketing campaign (categorical)

Output Variable

• target: Whether the client subscribed to a term deposit? (binary: "yes", "no")

Project Workflow

1. Exploratory Data Analysis (EDA)

   • Univariate, bivariate, and multivariate analysis using visualizations.
   • Correlation heatmap for feature selection.

2. Data Preprocessing

   • Handling missing values using SimpleImputer.
   • Encoding categorical variables with OrdinalEncoder and OneHotEncoder.
   • Scaling numerical variables using StandardScaler.

3. Feature Engineering

   • Derived new features like age_group, balance_group, and interaction terms.
   • Calculated derived metrics like campaign_intensity and contact_rate.

4. Model Building

   • Models: Logistic Regression, Random Forest, and XGBoost.
   • Hyperparameter tuning for optimal performance.
   • Metrics: Accuracy, Precision, Recall, F1-score, and AUC-ROC.

5. Results Visualization

   • Bar charts comparing model performance on test accuracy and F1-score.

6. Submission Generation

   • Predictions exported in the required format (submission.csv).

How to Use

1. Clone the repository.

   git clone https://github.com/username/banking-marketing-prediction.git

2. Install dependencies.

   pip install -r requirements.txt

3. Run the analysis script.

   python main.py

4. Generate submission files.

   • Logistic Regression: submission_lr.csv
   • Random Forest: submission_rf.csv
   • XGBoost: final_submission.csv

Results

Model	Train Accuracy	Test Accuracy	Train F1-Score	Test F1-Score
Logistic Regression	85.4%	83.2%	81.2%	80.1%
Random Forest	89.7%	87.5%	85.9%	84.3%
XGBoost	91.2%	88.7%	87.6%	86.2%

Visualizations

• Test Accuracy Comparison

• Test F1-Score Comparison

Future Improvements

• Feature selection optimization.
• Incorporate additional models for comparison.
• Implement model explainability techniques.

License

This project is licensed under the MIT License - see the LICENSE file for details.

---