Multi-cloud production ML system for HR attrition prediction with full drift monitoring, improving retention accuracy by 20%.
The Employee Attrition Prediction system is a production-grade ML pipeline that predicts which employees are at risk of leaving the organization, enabling HR teams to take proactive retention actions.
Deployed simultaneously on GCP, Azure, AWS, and Hugging Face, this project demonstrates true multi-cloud production deployment. The system integrates Evidently AI for continuous model drift monitoring — automatically triggering retraining when statistical distribution shifts are detected in incoming employee data.
The full MLOps stack — Docker, DVC, and MLflow — ensures every model version, dataset, and experiment is fully reproducible. A REST API enables real-time integration with HR information systems (HRIS) for live risk scoring of employee profiles.
Cleaning HR datasets, handling class imbalance with SMOTE, encoding categorical features for attrition factors.
Comparing Logistic Regression, Random Forest, and XGBoost for attrition classification. SHAP for explainability.
Logging parameters, metrics, and artifacts. Using Model Registry for version control and staging.
Packaging the model serving API in Docker. Writing multi-stage Dockerfiles for production images.
Configuring CI/CD pipelines for simultaneous deployment to GCP Cloud Run, Azure Container Apps, AWS ECS, and Hugging Face Spaces.
Setting up Evidently AI dashboards for data drift, concept drift, and model performance monitoring with automated alerts.