Flight Delay
Forecasting Model

Project Overview

This project builds a machine learning model to predict commercial flight delays using the Bureau of Transportation Statistics (BTS) On-Time Performance dataset — one of the most comprehensive publicly available sources of U.S. domestic flight data.

The model identifies which combination of factors — carrier, route, season, departure time, and weather category — most reliably predicts a delay, and by how much. Results are presented with feature importance reporting so the drivers of delay are clearly communicated alongside the predictions themselves.

Methodology

The project uses a supervised regression approach with the following pipeline:

• Data sourced from BTS On-Time Performance database — monthly CSVs covering U.S. domestic carriers
• Feature engineering applied to extract departure hour, day-of-week, season, and route-level delay history
• Weather category joined to flight records using FAA weather delay classifications
• Baseline linear regression model established, then compared against a Random Forest regressor for accuracy
• Feature importance extracted from the Random Forest model to rank delay predictors
• Model performance evaluated using RMSE and MAE across carrier and route segments
• Power BI dashboard built to visualize predicted vs. actual delay by route, carrier, and season

Dashboard & Charts

Power BI dashboard and model output charts will be embedded here upon project completion.

Key Findings

Findings will be documented here upon model completion. Expected outputs include:

• Top predictors of flight delay ranked by feature importance
• Routes and carriers with the highest predicted delay risk
• Seasonal and time-of-day patterns in delay probability
• Model accuracy metrics: RMSE and MAE by carrier segment
• Comparison of linear regression vs. Random Forest performance

Source Code

Full Python analysis code, model training pipeline, and documentation available on GitHub.