Understanding Aviation Sustainability with XAI
Presentation Description/Abstract
This project uses open aviation operations datasets to build machine-learning and explainable AI (XAI) models that forecast short- and medium-term sustainability outcomes. Target variables include gate-to-gate CO₂ estimates, flight efficiency (horizontal/vertical), surface inefficiency (e.g., additional taxi time), and delay-related indicators at network, state, airport, and service-provider levels. ML and XAI algorithm are trained with rolling-origin validation; performance is reported via SHAP values, partial-dependence plots, and counterfactual analyses provide transparent driver attribution and reveal leverage points—such as continuous climb/descent procedures, pushback control, and demand-capacity alignment. The outcome is a practical, interpretable toolkit that (i) predicts sustainability metrics with actionable confidence intervals, (ii) ranks interventions by marginal emissions-reduction potential, and (iii) supports data-driven prioritization of operational improvements across the aviation network.
Understanding Aviation Sustainability with XAI
Student Union Event Center
This project uses open aviation operations datasets to build machine-learning and explainable AI (XAI) models that forecast short- and medium-term sustainability outcomes. Target variables include gate-to-gate CO₂ estimates, flight efficiency (horizontal/vertical), surface inefficiency (e.g., additional taxi time), and delay-related indicators at network, state, airport, and service-provider levels. ML and XAI algorithm are trained with rolling-origin validation; performance is reported via SHAP values, partial-dependence plots, and counterfactual analyses provide transparent driver attribution and reveal leverage points—such as continuous climb/descent procedures, pushback control, and demand-capacity alignment. The outcome is a practical, interpretable toolkit that (i) predicts sustainability metrics with actionable confidence intervals, (ii) ranks interventions by marginal emissions-reduction potential, and (iii) supports data-driven prioritization of operational improvements across the aviation network.