Portuguese Bank Marketing Campaign Analysis

Project Overview

This project analyzes direct phone marketing campaigns by a Portuguese bank (2008-2010) to predict term deposit subscriptions. It combines exploratory data analysis, predictive modeling, and strategic recommendations to optimize customer targeting. The workflow includes:

1. Data Analysis: Exploratory insights into customer demographics, economic indicators, and campaign performance.
2. Predictive Modeling: Development of machine learning models (LightGBM, XGBoost, RandomForest) to identify high-potential customers.
3. Actionable Recommendations: Data-driven strategies to enhance campaign effectiveness.

Dataset

Domain: Banking/Finance
Source: UCI Machine Learning Repository
Enriched Data: Includes macroeconomic indicators from Banco de Portugal.
Size: 41,188 records × 21 features.

Key Features

• Demographics: Age, job, marital status, education.
• Financial: Credit default, housing/personal loans.
• Campaign Metrics: Contact type, duration, month/day of contact.
• Economic Indicators: Euribor rate, employment rate, consumer confidence index.
• Target: y (binary: "yes"/"no").

Dataset Link: Download Here

---

Repository Structure

├── data/                 # Raw and processed datasets
├── docs/                # Project briefs and dataset documentation
├── notebook/            # Jupyter notebook for analysis and modeling
├── report/              # Final report and strategic suggestions
├── results/             # Visualizations, model outputs, and analysis reports
├── scripts/             # Utility scripts and helper functions
└── requirements.txt     # Python dependencies

Installation

1. Clone the Repository:

   git clone https://github.com/dhaneshbb/ProtugeseBank.git
   cd ProtugeseBank

2. Install Dependencies:

   pip install -r requirements.txt

3. Custom Library: Install insightfulpy for streamlined analytics:

   pip install insightfulpy

---

Usage

Running the Analysis

1. Open the Jupyter notebook:
   notebook/PRCP-1000-PortugeseBank.ipynb
2. Execute cells sequentially to:
   • Perform data cleaning and EDA.
   • Train and evaluate predictive models.
   • Generate visualizations and reports.

Key Outputs

• Final Model: results/model/final_lgbm_model.pkl (LightGBM with 88.8% accuracy).
• Reports:
  • report/Final Report.md: Comprehensive analysis and recommendations.
  • results/365csv pre-anlysis/: Preprocessing reports, visualizations, and statistical summaries.

---

Model Performance

Model	Accuracy	Precision	Recall	F1-Score	ROC AUC
LightGBM	88.8%	47.1%	57.3%	51.5%	81.2%
XGBoost	87.7%	46.4%	58.7%	51.9%	80.9%
RandomForest	86.8%	43.8%	60.2%	50.7%	79.7%

Challenges Faced

1. Data Imbalance: Addressed using class weighting and LightGBM’s is_unbalance parameter.
2. Multicollinearity: reduced complexity Mitigated via VIF analysis and tree-based models.
3. Overfitting: Controlled through cross-validation and regularization.

---

References

Moro, S., Cortez, P., & Rita, P. (2014). A Data-Driven Approach to Predict the Success of Bank Telemarketing. Decision Support Systems, In press. DOI: 10.1016/j.dss.2014.03.001.

• Available at:
  • PDF
  • BIB

---

License: This project is for educational/research purposes. Cite the dataset authors when referencing this work.