AI in Aviation Case Study: Predicting Taxi Times

by Dr. Phil Winder , CEO

Our client, a global aviation business continually seeks innovative ways to enhance airline efficiency and reduce operational disruptions. Ground operations, particularly taxi-out times, have been a persistent source of delays and unplanned costs. By collaborating with Winder.AI, our client aimed to build a robust, machine learning (ML) solution capable of predicting taxi times at various time horizons, leading to improved scheduling, minimized gate-holding, and reduced fuel consumption.

Our client now offers a Taxi-Time API to share these insights with industry stakeholders. The overarching goal is to give airlines a predictive edge, helping them streamline ground operations and cut down on costly inefficiencies.

Motivation for the Taxi-Time API

According to publicly available information, the Taxi-Time API addresses several key industry challenges:

  1. Operational Planning
    • Airline dispatchers often struggle to accurately schedule departure and arrival gates due to unknown taxi-out durations. By improving predictions, airlines can better align ground crew tasks and optimize resource allocation.
  2. Reducing Fuel Consumption
    • Prolonged idling on taxiways contributes significantly to unnecessary fuel burn. Better insights enable pilots and operators to manage engine usage and gate departure more strategically.
  3. Improving On-Time Performance
    • Underestimating or overestimating taxi times can disrupt entire flight schedules. More precise data supports advanced warning systems, allowing airlines to plan contingencies or reallocate resources effectively.
  4. Passenger Satisfaction & Safety
    • Delays on the ground translate to missed connections, heightened stress, and potential safety risks when schedules become too tight. Proactive planning through predictive analytics enhances the passenger experience.

Winder.AI played a pivotal role in strengthening these objectives by researching and improving an ML model that powers the core predictive capabilities of the Taxi-Time solution.

Challenges & Objectives

During development of the taxi-time prediction solution, the team had to contend with data drawn from multiple and often disparate sources. Certain operational details, such as last-minute runway configuration changes or infrequent aircraft maintenance events, are not consistently logged, making it difficult to capture every factor that influences ground delays. Additionally, the initial cloud platform used for building and testing the machine learning models offered limited flexibility around custom monitoring, prompting a shift to an environment with more comprehensive oversight and analytics capabilities.

The project’s primary focus was on improving a predictive model that could accurately forecast taxi-out durations. This meant improving mean absolute error (MAE) over existing approaches and increasing the number of flights predicted within a tighter time window, generally measured in seconds.

Another essential goal was ensuring these predictions could be generated far enough in advance to guide meaningful operational decisions such as gate assignments, crew scheduling, and fuel planning. Through continuous experimentation and refinements, the team was able to meet these objectives and improve the models used in the Taxi-Time API.

Approach & Methodology

Data Collection & Exploration

  • Scope
    Processed hundreds of thousands of flight records incorporating a wide variety of features.
  • Stand-to-Runway Focus
    Winder.AI identified that taxi routes strongly influence total taxi-out time. Factoring in each airport’s unique layout was essential.

Model Development

  • Independent Analysis
    Winder.AI operated in a “blind” manner, without leveraging our client’s previous work, ensuring a fresh perspective.
  • Feature Engineering
    Incorporated a range of new features into the model.
  • Iterative Experimentation
    Ran extensive experiments, refining the machine learning pipeline to minimize errors and robustly handle incomplete data.

Deployment & Monitoring

  • Platform Decision
    After outgrowing the limited metrics customization of the first ML platform, Winder.AI helped the team to develop a more comprehensive MLOps environment, capable of detailed performance tracking (e.g., feature drift, prediction drift).

Results & Impact

  1. Improved Accuracy & Reliability
    • Lower MAE than existing baselines, outperforming earlier internal benchmarks.
    • Increased on-target predictions within ±300 seconds, ensuring operational decisions are more informed and proactive.
  2. Operational & Financial Benefits
    • Reduced Fuel Burn: Airlines can optimize engine start times and mitigate taxiway congestion.
    • Minimized Gate-Holding: More reliable estimates help gate management teams plan turnarounds efficiently.
    • Enhanced On-Time Performance: Potentially fewer delays cascading through the day’s flight schedule.
  3. Foundation for a Live API
    • The resulting production-ready model is being integrated into our client’s next-generation Taxi-Time API.
  4. Scalable & Future-Ready
    • Built to handle high volumes of incoming requests and new data feeds.
    • Customizable architecture allows our client to integrate emerging data sources or adapt the model for additional airports worldwide.

Key Takeaways

  1. New Insights from Pre-Flight Data
    Winder.AI demonstrated that long-term forecasting (days or even months in advance) can closely match near-real-time forecasts, empowering far more strategic planning than was previously possible.
  2. Innovation Under Organizational Shifts
    The team navigated organizational change, highlighting Winder.AI’s adaptability and commitment to delivering business value.
  3. End-to-End Solution
    From initial experimentation through to final deployment, the approach leveraged and improved our client’s workflows and improved an enterprise-scale model-monitoring framework.
  4. Driving Tangible Airline Benefits
    The Taxi-Time API now provides actionable intelligence for airlines worldwide, showcasing the benefits of digitizing aviation operations and reducing operational inefficiencies.

Conclusion & Call to Action

The quest to improve ground operations and deliver advanced predictive services is exemplified by the Taxi-Time API. By partnering with Winder.AI, our client overcame complex data challenges, achieved meaningful accuracy gains, and introduced a live, production-grade service for industry stakeholders.

If your organization faces operational forecasting or efficiency challenges in aviation, logistics, or another industry, Winder.AI stands ready to help.

Take the Next Step

  • Explore an example taxi-time API by searching for a taxi-time developer portal.
  • Visit Winder.AI to discover our capabilities in machine learning, data engineering, and digital transformation.

More articles

Enterprise AI Assistants: Combatting Fragmentation

Data silos are a cause of fragmentation in the enterprise. But what if we could use AI Assistants to connect them?

Read more

Scaling StableAudio.com Generative Models Globally with NVIDIA Triton & Sagemaker

Learn from the trials and tribulations of scaling audio diffusion models with NVIDIA's Triton Inference Server and AWS Sagemaker.

Read more
}