banner letter Contact Us


Fast break to efficiency

Managing carry-on luggage efficiently is pivotal to streamlining boarding in the airline industry. Gate agents face pressure to plan effectively during pre-flight meetings, where accurately estimating the number of carry-on bags is crucial. This estimation not only aids in managing cabin space to ensure adequate overhead bin availability for passengers’ belongings but also informs strategies for allocating boarding time and expediting the process in cases of higher baggage volumes.

A reliable solution to seamlessly predict carry-on bag numbers has become increasingly essential to enhance these predictions’ accuracy and improve the overall efficiency of the boarding process. While this challenge is most prominent in airlines, similar issues are pertinent in other travel industry sectors, such as hotels and cruise lines, where baggage management impacts operational efficiency.


Tackling the turnover in baggage predictions

Since 2016, our client has grappled with inaccurate carry-on bag predictions, with less than 20% of their estimates proving reliable. Their existing model consistently underpredicted gate bags for flights, leading to inefficient baggage handling and frequent flight delays.

Data availability

The need for systematic integration of various data sources, including checked baggage records and flight data, and the absence of real-time bin space monitoring hindered effective carry-on baggage management. These shortcomings, along with premature evaluations by gate agents and a lack of real-time data and feedback, made informed decision-making difficult.

Behavioral aspects

The existing baggage prediction system faced low adoption among gate agents due to its lack of explainable predictions and context, resulting in diminished trust. Additionally, frequent prediction errors further exacerbated agent dissatisfaction, reducing their confidence in relying on the system for effective baggage management.

Model flexibility

The rigid batch deployment approach led to an inflexible system architecture, making model tuning and updates laborious and time-consuming. A flexible, real-time approach was essential for effectively predicting and managing airline baggage.


Crafting a winning strategy

The solution required a systematic and multifaceted approach. It integrated diverse data types, including checked baggage, historical and future flight schedules, active bookings, and passenger details. Building predictive models was crucial, with a focus on incorporating real-time information.

Feature engineering was a crucial component in this process, enhancing model accuracy by identifying essential features like the number of days before holidays and load factors. The model also needed to consider the behavioral patterns of gate agents, the number of gate-checked bags from previous flights, and the count from the previous week’s flight on the same day for the same market.

What we provided:

Gameplay with predictive analytics

Predictive modeling techniques Infrastructure and deployment Collaborative validation and improvement

Applied CatBoost Regression to create a decision tree ensemble for accurate predictions

Developed on AWS with a focus on a flexible architecture to facilitate continuous updates

We worked closely with the data science community and airline gate agents for real-world validation

Analyzed BE Load factor, Directional market, Load Factor, Equipment type, and % Single Pax PNRs for predicting gate bags

implemented an API-driven model-as-a-service for timely data updates and retraining

Developed a user-friendly front-end application displaying prediction ranges to enhance agent trust

Conducted experiments with oversampling and under-sampling techniques, leveraging the BE Load factor for significant accuracy improvements

Took 6–9 months for design and development, with an additional 3 months for implementation

Continuous efforts in integrating end-user feedback to refine the process and methodology


Netting success with enhanced baggage management

The airline transformed its approach to managing carry-on baggage by harnessing cutting-edge machine-learning techniques and meticulous data analysis. This solution not only refined the accuracy of baggage predictions but also streamlined the overall baggage handling process, marking a significant advancement in enhancing the client’s operational efficiency and passenger satisfaction.

The immediate impact:

By operationalizing the solution, the client improved the accuracy of the existing machine-learning model to forecast the number of gate-checked bags. The development of a feature store on the client’s modern deployment infrastructure streamlined the architecture, facilitating more accessible model updates and leading to increased adoption by gate agents, substantially reducing overall flight delays.


Accuracy achieved

Increased agent adoption

Decreased onboarding delays
Sustainable Gains

In the long term, the client will be able to pinpoint high-risk categories and stores, enabling better resource allocation, which, in turn, is expected to lead to a ~10-20% reduction in shrinkage within just one year.


Let’s collaborate!

    *By submitting your details, you are agreeing to receive communications about Fractal resources, events, products, or services. By submitting this form you are acknowledging that you have read and agree to our privacy statement.