ML/AI

Churn Prediction Model

Scikit-learn pipeline predicting member churn with 89% accuracy.

Client
Internal Project
Industry
Machine Learning / Healthcare
Duration
2 months (2024)
Role
ML Engineer
Churn Prediction Model

Overview

Developed a machine learning pipeline to predict member churn using historical activity data, enabling proactive retention strategies.

Key Results

89%
Accuracy
Achieved
50+
Features
Engineered
<100ms
Latency
Per Prediction
30%
Churn
Reduction

The Challenge

The healthcare platform was losing members without early warning:

  • No visibility into churn risk factors
  • Reactive approach to member retention
  • Manual analysis of member behavior
  • Inconsistent intervention strategies
  • High false-positive rates in existing rules

They needed a data-driven approach to predict and prevent churn.

The Solution

I built an end-to-end ML pipeline:

Feature Engineering

Extracted 50+ features from activity logs, plan usage, and engagement data.

Model Training

Tested multiple algorithms (Random Forest, XGBoost, Logistic Regression).

API Deployment

FastAPI endpoint for real-time predictions.

Monitoring

Model performance tracking and drift detection.

Key Features

🎯

89% Accuracy

High-precision predictions with low false positives.

âš¡

Real-Time

Sub-100ms prediction latency.

📊

Explainable

SHAP values for feature importance.

🔄

Auto-Retrain

Scheduled model updates with new data.

Tech Stack

ML Framework

Scikit-learnXGBoostPandas

API

FastAPIPydantic

Data

PostgreSQLApache Airflow

MLOps

MLflowDockerAWS SageMaker

Screenshots

Model

Model Performance

Features

Feature Importance

Dashboard

Prediction Dashboard

Achievements

  • Achieved 89% prediction accuracy
  • Reduced churn by 30% through early intervention
  • Sub-100ms prediction latency
  • Automated weekly model retraining