Bank Term Deposit Subscription Prediction

İsmet Erdal Tunç & Ozan Tekin

Economic Question & Motivation

Can client demographic, financial, and campaign-related characteristics predict whether a client subscribes to a bank term deposit?

Why is this important?

  • Helps banks target customers more effectively
  • Improves marketing efficiency
  • Supports data-driven decision making

Dataset Description

Dataset: Bank Marketing Dataset

Source:

- Portuguese Retail Bank

- UCI Repository (Kaggle Mirror)

- 2008–2010

Observations: 11,162 clients

Outcome Variable:

- Deposit (Yes / No)

Key Predictors: Balance, Housing Loan, Personal Loan, Duration, Previous Campaign Outcome

Probability & Distribution Analysis



- 52.6% No
- 47.4% Yes
- Balanced dataset

Modeling Strategy

We built two logistic regression models.

Model 1

Financial Variables: - Balance - Housing Loan - Personal Loan

Model 2

Financial + Campaign Variables: - Balance - Housing Loan - Personal Loan - Duration - Previous - Previous Campaign Outcome (poutcome)

Goal

Determine whether campaign-related variables improve predictive performance.

Model Comparison

Model Accuracy Precision Recall
Model 1 0.603 0.627 0.626
Model 2 0.789 0.770 0.861

Key Finding

Model 2 outperformed Model 1 across all evaluation metrics.

Cross Validation

Metric Test Set Cross Validation
Accuracy 0.789 0.786
Precision 0.770 0.769
Recall 0.861 0.846

Key Finding

The cross-validation results are very similar to the test-set results.

Interpretation: Model 2 generalizes well and shows no strong evidence of overfitting.

Economic Interpretation

↑ Balance

↑ Duration

↑ Previous Success

↓ Housing Loan

↓ Personal Loan

→ Better customer targeting

Limitations & Future Work

Limitations

  • Missing income & wealth variables
  • Single banking dataset

Future Work

  • Random Forest
  • Gradient Boosting
  • Customer loyalty analysis

Conclusion

  • Best Model: Model 2
  • Accuracy: 78.9%
  • Recall: 86.1%
  • Stable across CV
  • Better customer targeting