Flight delays can be very frustrating to passengers and costly to airline companies. Flight delays are not easy to understand as they arise from multiple reasons like increase in air traffic at the origin or destination airport, weather etc. The on-time performance data of airline schedules could be useful to shed some light into causes of flight delays.
The goal of this project is to construct a flight delay prediction model leveraging flight On-Time performance data and use the model for deriving insights about past flight delays. The on-time arrival data for non-stop domestic flights from US Department of Transportation(DoT) was used for this analysis.
Essentially, my modeling approach involves using traditional & popular machine learning algorithms like Logistic Regression, Decision trees & Random Forest to predict airline delays This notebook contains sections for dataset exploration, data cleansing, deriving insights from visualizations, training and evaluation of classification models, and model interpretation. Common model evaluation metrics like sensitivity, specificity, ROC curves and accuracy scores were used for model evaluation.