Dynamic Customer Churn Prediction Engine

The sales and customer success teams were operating reactively, often only addressing customer dissatisfaction after a cancellation request was filed. They needed a way to identify "at-risk" accounts early and—more importantly—understand why those specific customers were likely to leave so they could offer targeted interventions.

I built and deployed a predictive analytics application that transforms raw customer data into a real-time risk assessment. Unlike a static monthly report, this solution features an interactive "What-If" simulation interface, allowing stakeholders to test how changes in pricing or service levels would impact retention probabilities.

Modeling: Developed a Random Forest Classifier using scikit-learn, trained on historical customer telemetry (contract tenure, monthly spend, and support ticket volume).

Feature Engineering: Identified "Tech Support Interactions" and "Monthly Recurring Revenue (MRR)" as the two highest-weighted features for churn prediction.

Application Layer: Wrapped the model in a Streamlit web interface to democratize access for non-technical stakeholders.

Performance: The model achieved an F1-score of 0.82 on the test set, balancing precision and recall to minimize false alarms while catching the majority of at-risk accounts.

Core: Python 3.9, Pandas, NumPy

ML: Scikit-learn (Random Forest), Joblib (Model Serialization)

UI/Deployment: Streamlit

Architecture: RESTful design pattern for inferencing

shifted the retention strategy from reactive to proactive. The dashboard empowers the sales team to visualize the direct correlation between support availability and customer longevity, providing data-backed justification for resource allocation.