Predicting aviation KPIs with over 93% accuracy using real BTS datasets, end-to-end MLOps pipelines, and multi-cloud deployment.
The AirTransport Predictor is a production-grade Machine Learning system designed to forecast critical aviation Key Performance Indicators — including passenger volumes, freight tonnage, and mail loads — using real-world datasets from the Bureau of Transportation Statistics (BTS) of the United States.
The system achieves over 93% predictive accuracy through rigorous feature engineering, model selection, and hyperparameter optimization. The entire pipeline is automated using MLOps best practices: DVC manages data versioning, MLflow tracks all experiments and model artifacts, and GitHub Actions orchestrates CI/CD deployment to AWS.
A Streamlit dashboard provides real-time KPI visualization and forecasting, while a REST API allows integration with external operational systems. The project demonstrates how industrial-scale ML can be built, validated, monitored, and updated entirely in an automated fashion.
Downloading and validating BTS aviation datasets. Setting up DVC remote storage and pipeline stages.
Time-series feature extraction, lag variables, rolling statistics, and categorical encoding for aviation data.
Training and comparing Random Forest, Gradient Boosting, and XGBoost models. MLflow experiment logging.
Building a Flask/FastAPI REST endpoint for real-time KPI prediction with input validation.
Interactive Streamlit app for exploring forecasts, confidence intervals, and historical trends.
Dockerizing the app, configuring GitHub Actions workflow, and deploying to AWS EC2/ECS with automated rollback.