https://github.com/udityamerit/customer-segmentation-and-churn-prediction-model
https://github.com/udityamerit/customer-segmentation-and-churn-prediction-model
machine-learning machine-learning-algorithms scikit-learn scikitlearn-machine-learning
Last synced: 7 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/udityamerit/customer-segmentation-and-churn-prediction-model
- Owner: udityamerit
- License: mit
- Created: 2024-11-27T15:58:26.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2024-11-28T17:14:13.000Z (10 months ago)
- Last Synced: 2025-01-27T06:43:00.103Z (8 months ago)
- Topics: machine-learning, machine-learning-algorithms, scikit-learn, scikitlearn-machine-learning
- Language: Python
- Homepage:
- Size: 154 KB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
## **Customer Segmentation and Churn Prediction Model**
### **Project Overview**
This project aims to:
1. Segment customers into clusters based on their demographic, spending, and engagement patterns.
2. Predict customer churn to help businesses focus their marketing efforts on retaining at-risk customers.
The app is built with **Streamlit**, using clustering (KMeans) and machine learning techniques for churn prediction. The model provides actionable insights to improve customer engagement and retention strategies.
---
### **Features**
1. **Dynamic Clustering**:
- Users can choose the number of clusters (2 to 10) for segmentation.
- Clustering helps group customers with similar characteristics for targeted campaigns.
2. **Churn Prediction**:
- Key features like spending habits, age, income, and purchase frequency are used to predict churn probabilities.
3. **Visualization**:
- PCA-based scatter plots to visualize customer segments.
- Histograms for feature distribution across clusters.
4. **Marketing Strategy Recommendations**:
- Tailored recommendations for each customer cluster.
---
### **How to Run**
1. Install dependencies:
```bash
pip install streamlit pandas numpy matplotlib seaborn scikit-learn
```
2. Save the script as `customer segmentation.py`.
3. Run the app:
```bash
streamlit run customer segmentation.py
```
4. Upload your dataset (`.csv` file) and explore clusters or churn predictions.
---
### **Model Description**
1. **Clustering**:
- Features used: Age, Income, Spending, Purchases, Campaign Response Rates, etc.
- Preprocessing includes missing value handling, scaling, and one-hot encoding.
- PCA for dimensionality reduction and visualization.
2. **Churn Prediction**:
- Supervised learning techniques (e.g., Logistic Regression, Random Forest) can be integrated for churn prediction.
- Important features include customer spending, campaign response rates, and frequency of purchases.
---
### **Answers to Key Questions**
#### **1. How would you handle imbalanced data if churned customers are fewer than active ones?**
- **Oversampling**: Use techniques like SMOTE (Synthetic Minority Oversampling Technique) to generate synthetic samples of the minority class.
- **Undersampling**: Randomly sample from the majority class to balance the dataset.
- **Class Weighting**: Modify the loss function of the model to penalize misclassifications of the minority class more heavily.
- **Evaluation Metrics**: Use metrics like F1-score, precision, recall, and ROC-AUC instead of accuracy to evaluate the model performance on imbalanced data.
#### **2. What features are the most important predictors of churn?**
- **Total Spending**: High spending may correlate with loyalty, while erratic spending patterns may indicate potential churn.
- **Campaign Response Rate**: Customers who do not respond to marketing campaigns are more likely to churn.
- **Number of Purchases**: Lower purchase frequency can be an early indicator of churn.
- **Income**: Disposable income impacts purchasing behavior, which may relate to churn risk.
- **Web Visits**: High web visits without purchases may indicate dissatisfaction or exploration of alternatives.
#### **3. How would you explain the model's predictions to a non-technical business team?**
- **Cluster Analysis**: Use visualizations (e.g., PCA plots) to show how customers are grouped based on similar characteristics.
- **Key Metrics**: Present insights such as average age, spending, and income per cluster in simple terms.
- **Churn Insights**: Explain churn probabilities using concrete examples, such as "Customers who spend less than $50 per month and do not respond to campaigns are 3x more likely to churn."
- **Business Impact**: Highlight how retaining customers in high-risk clusters can increase revenue and customer lifetime value.
#### **4. What steps would you take to deploy this model into production?**
1. **Model Packaging**:
- Save the trained model using a format like `pickle` or `joblib`.
- Create a robust pipeline for preprocessing and predictions.
2. **API Development**:
- Use **FastAPI** or **Flask** to expose the model as an API endpoint.
- Ensure secure and scalable API endpoints.
3. **Integration**:
- Embed the API in the existing CRM or marketing platforms.
- Ensure the application can handle real-time inputs and predictions.
4. **Monitoring**:
- Track model performance using dashboards (e.g., Grafana).
- Monitor drift in data distribution and retrain the model as needed.
5. **Feedback Loop**:
- Use feedback from business teams and customers to iteratively improve the model.
---
### **Folder Structure**
```
📂 Customer-Segmentation
├── 📄 customer segmentation.py
├── 📄 README.md
└── 📂 data
└── marketing_campaign.csv
```