🏆 MLOps Project: Bank Marketing Prediction API

---

📢 About

This project was completed as part of the Veri Bilimi Okulu MLOps Bootcamp.
It showcases the end-to-end implementation of a machine learning pipeline, deployment with FastAPI, Docker containerization, and infrastructure automation using Terraform and Ansible on AWS.

---

📌 Problem Statement

Financial institutions often run marketing campaigns to promote term deposit products.
The goal of this project is to predict whether a customer will subscribe to a term deposit using historical campaign data.

This project demonstrates an end-to-end MLOps workflow by building a machine learning pipeline, deploying it via FastAPI, containerizing with Docker, and provisioning infrastructure using Terraform and Ansible on AWS EC2.

---

📊 Dataset

• Source: UCI Machine Learning Repository – Bank Marketing Dataset
• File: bank-additional-full.csv
• Samples: 41,188
• Features: 21
• Target: y – whether the client subscribed to a term deposit (yes / no)

---

⚙️ Model Pipeline

🔧 Preprocessing

• Label encoding for categorical variables

🧠 Feature Engineering

• contacted_before: derived from pdays
• age_group: bucketized from age
• Removed multicollinearity (via correlation matrix + VIF)
• Feature selection prior to scaling
• Scaling with RobustScaler

🧮 Modeling

• VotingClassifier ensemble with:
  • XGBoostClassifier (use_label_encoder=False, eval_metric='logloss')
  • CatBoostClassifier (silent mode)
  • RandomForestClassifier (n_estimators=100)
• Evaluated using Stratified 5-Fold Cross-Validation
• Best model saved as voting_model.pkl

🔢 Features Used

['job', 'marital', 'education', 'default', 'housing', 'loan', 'contact', 'month',
 'day_of_week', 'campaign', 'pdays', 'previous', 'poutcome', 'cons.conf.idx',
 'nr.employed', 'age_group']

---

🧪 Running Locally

1. Build Docker Image

docker build -t mlops-fastapi -f docker/Dockerfile .

2. Run the API Container

docker run -p 8000:8000 mlops-fastapi

3. Test Prediction

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d @api/test_payload.txt

4. Open Swagger UI

Open your browser and navigate to: http://localhost:8000/docs

---

☁️ Deploy to AWS EC2

1. Provision EC2 Instance with Terraform

cd infra/terraform
terraform init
terraform apply

2. Deploy App with Ansible

cd ../ansible
ansible-playbook -i inventory playbook.yml

---

📨 Example Request Payload

{
  "job": "technician",
  "marital": "single",
  "education": "university.degree",
  "default": "no",
  "housing": "yes",
  "loan": "no",
  "contact": "cellular",
  "month": "may",
  "day_of_week": "mon",
  "campaign": 1,
  "pdays": -1,
  "previous": 0,
  "poutcome": "nonexistent",
  "cons_conf_idx": -36.4,
  "nr_employed": 5191.0,
  "age_group": "25-35"
}

---

🔍 Drift Detection (Future Scope)

Although drift detection is not implemented in this version, the project is structured for easy integration of drift detection techniques, such as:

• Statistical monitoring of feature distribution (e.g., using EvidentlyAI)
• Performance monitoring via accuracy/AUC tracking
• Triggering retraining when significant drift is detected

---

📁 Directory Structure

MLOPS_PROJECT/
├── api/              # FastAPI implementation
│   ├── api_main.py
│   ├── predict.py
│   └── test_payload.txt
├── docker/           # Dockerfile
│   └── Dockerfile
├── infra/            # Infrastructure as code
│   ├── terraform/
│   └── ansible/
├── ml_model/         # ML logic and training pipeline
│   ├── train_model.py
│   ├── preprocess.py
│   ├── main.py
│   ├── EDA.py
│   ├── utils.py
│   ├── bank-additional-full.csv
│   └── saved_models/
├── requirements.txt  # Dependencies

---

👩‍💻 Maintainer

Project by: Berna Yılmaz
📧 berna14y@gmail.com

---

✅ Summary

This project demonstrates a complete MLOps lifecycle:
From preprocessing and modeling to containerized deployment on AWS infrastructure.
It provides a fast, stateless prediction service and is extensible for further automation, CI/CD, and monitoring.

---