How deuellau/bank-telemarketing-campaign works
This repository is a bank telemarketing prediction project, documented primarily through a Jupyter notebook. Its workflow takes campaign data through leakage-aware feature preparation, trains a stacked model, and uses predicted subscription likelihoods to prioritize calls and estimate campaign economics. The README documents the model components, nested cross-validation, and business results; no notebook source was available to verify finer-grained implementation details, so the graph shows only documented relationships.
Main entry points: Project Workflow.ipynb Subscription scores
Components 9
Data preparation
- Feature preparation
Code/Project Workflow.ipynb
Modeling
- Base learners
Code/Project Workflow.ipynb - OOF predictions
Code/Project Workflow.ipynb - Logistic meta-classifier
Code/Project Workflow.ipynb - Nested cross-validation
Code/Project Workflow.ipynb
Campaign decisions
- Subscription scores
Code/Project Workflow.ipynb - Call prioritization
Code/Project Workflow.ipynb - Campaign economics
Code/Project Workflow.ipynb
Other components
- Campaign data
Connections 9
- Campaign data → Feature preparation: provides records
- Feature preparation → Base learners: provides features
- Base learners → OOF predictions: generate
- OOF predictions → Logistic meta-classifier: train
- Nested cross-validation → Base learners: tunes and evaluates
- Nested cross-validation → Logistic meta-classifier: evaluates
- Logistic meta-classifier → Subscription scores: produces
- Subscription scores → Call prioritization: ranks leads
- Subscription scores → Campaign economics: supports threshold analysis