https://github.com/thedatatenno/churn-prediction
A machine learning project to predict customer churn using classification models with SMOTE and hyperparameter tuning."
https://github.com/thedatatenno/churn-prediction
accuracy-score churn-prediction classification confusion-matrix data-science f1-score gridsearchcv knn-classifier logistic-regression machine-learning matplotlib portfolio-project python random-forest scikit-learn seaborn smote telecom xgboost-classifier
Last synced: 11 months ago
JSON representation
A machine learning project to predict customer churn using classification models with SMOTE and hyperparameter tuning."
- Host: GitHub
- URL: https://github.com/thedatatenno/churn-prediction
- Owner: TheDataTenno
- License: mit
- Created: 2025-04-06T03:55:17.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-04-06T05:12:49.000Z (over 1 year ago)
- Last Synced: 2025-04-09T22:12:31.955Z (over 1 year ago)
- Topics: accuracy-score, churn-prediction, classification, confusion-matrix, data-science, f1-score, gridsearchcv, knn-classifier, logistic-regression, machine-learning, matplotlib, portfolio-project, python, random-forest, scikit-learn, seaborn, smote, telecom, xgboost-classifier
- Language: Jupyter Notebook
- Homepage:
- Size: 1.73 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
📊 Customer Churn Prediction
This project predicts customer churn using a supervised machine learning approach for a telecom company. While this dataset comes from the telecom industry, the methodology and techniques demonstrated here are **scalable and adaptable to any industry** where customer retention is vital—such as finance, insurance, SaaS, e-commerce, and beyond.
This project serves as a **showcase of my data science skillset**, including data preprocessing, class imbalance handling, algorithm comparison, hyperparameter tuning, and model interpretation.
---
## 🔍 Problem Statement
Customer churn—when a customer discontinues service—is a key metric for subscription-based companies. For this project, we analyze customer behavior to predict churn using various customer and service attributes. Our goal is to **identify customers at risk of churning**, allowing businesses to take proactive retention actions.
---
## 🧠 Algorithms Used
- **Logistic Regression** (best performing model after SMOTE + tuning)
- **Random Forest** (tuned using GridSearchCV, competitive performance)
We experimented with **K-Nearest Neighbors (KNN)** and **XGBoost**, but deprioritized them based on evaluation results.
---
## 🧪 Data Preprocessing
- Removed irrelevant columns (`customerID`)
- Encoded categorical variables with `LabelEncoder`/`get_dummies`
- Addressed class imbalance using **SMOTE** (on training set only)
- Train/test split: 80% train / 20% test
---
## 📈 Model Evaluation
Metrics used to evaluate models:
- Accuracy
- Precision, Recall, F1-Score (especially for minority class)
- Confusion Matrix
- Feature Importance (visualized)
---
## 🔧 Best Model & Performance
**Best Model**: Logistic Regression
**Tuned Hyperparameters**:
```python
{ 'C': 0.1, 'penalty': 'l1', 'solver': 'liblinear' }
```
---
| Metric | Score |
|--------------|-------|
| Accuracy | 0.779 |
| Precision | 0.58 (for churned class) |
| Recall | 0.63 (for churned class) |
| F1-Score | 0.60 (for churned class) |
---
## 🔍 Feature Importance
Below are the top 10 most influential features based on the two best models:
### Logistic Regression
### Random Forest
---
## 🧰 Tech Stack
- **Languages**: Python 3.12
- **Libraries/Tools**:
- `pandas`, `numpy`, `matplotlib`, `seaborn` for data manipulation and visualization
- `scikit-learn` for model training, evaluation, and hyperparameter tuning
- `imblearn` for SMOTE-based class balancing
- `joblib` for model saving
- `Jupyter Notebook` + `VS Code` for development
---
## 👨🏽💻 Author
**Patrick Onuoha Jr.**
Data Science Senior @ University of North Texas
🔗 [GitHub](https://github.com/TheDataTenno) | 📧 [Email](mailto:ponuoha2017@gmail.com) | 🧠 Passionate about building data-driven solutions that scale across industries.