Inspiration:
In Rwanda and beyond, many businesses struggle with customer retention. Telecoms, banks, SaaS companies, and even schools often lose clients without clear early warnings. We were inspired to use Machine Learning to build a tool that predicts customer churn before it happens — allowing companies to act early, retain more customers, and grow sustainably.
What it does:
ChurnGuard is a machine learning-powered tool that analyzes customer behavior and predicts whether a customer is likely to leave a service. It takes customer data (such as subscription type, billing method, contract duration, and usage) and returns a prediction: "Likely to Churn" or "Not Likely to Churn". It also includes a simple dashboard (built with Streamlit) that helps businesses interactively test the model and understand key churn-driving factors.
Challenges we ran into:
Handling missing or inconsistent data during cleaning.
Dealing with imbalanced classes (fewer churners than non-churners).
Choosing the right features and avoiding overfitting.
Designing a simple but useful dashboard that’s both clear and functional.
Accomplishments that we're proud of:
Successfully trained a model with over 80% accuracy.
Built a working dashboard that demonstrates the model’s predictions in real-time.
Turned a real-world business problem into an explainable and scalable AI solution.
Learned best practices in data preprocessing, model evaluation, and deployment.
What we learned:
Practical applications of classification algorithms (Random Forest, Logistic Regression).
The importance of data quality and feature engineering in ML.
How to use Streamlit for building fast ML dashboards.
How to interpret model results using confusion matrix and performance metrics like precision, recall, and F1-score.
What's next for ChurnGuard AI:
Improve feature importance visualization to better explain model decisions.
Integrate with real-time business systems (e.g., CRM or SMS).
Add an alert system for sales/retention teams.
Train the model on local datasets from Rwanda to increase relevance and accuracy.
Package the app for production using Docker and CI/CD tools.
Log in or sign up for Devpost to join the conversation.