Cockpit Demand Forecasting

In order to determine the crew manpower requirements for the next 36 months, a comprehensive forecast of non-flight engagements for the flight crew is essential. This includes forecasting simulation and training sessions, resignations, annual leave, office and ground duties, unexpected losses, instructor duties, and P3 class duties. These forecasts will be broken down by fleet type, captain, and first officer (F/O) to ensure a detailed and accurate assessment of future needs. It is important to note that forecasting non-flight engagements is a critical part of crew demand prediction because these engagements are too unpredictable. For this reason we used artificial intelligence technics in this project.

Openshift container platform,

Python,

Oracle database,

Although the problem may seem like a classic time series problem, in addition to classic time series forecasting methods, the problem has been transformed into a regression problem, and both classic and advanced regression methods have been tested.

In this perspective, more than 1800 model trials have been conducted, and among them, the 150 models that yielded the best results are currently being used in production. Every month, the model processes new data by cleaning, transforming, modeling, selecting the best models, and predicting the next 36 months in a rolling window manner with new models. An autonomous machine learning model structure has been established, allowing the model to perform these tasks independently.

The crew requirement calculation work was being done manually. Despite being carried out manually, the ongoing work did not meet the current requirements. The biggest risk of the existing work arrangement was the excessive use of human resources in business units and the

On the other hand, non-flight engagements such as simulation and ground training, resignations, annual leave, office and ground duties, and instructor duties occur very spontaneously and unpredictably, making future crew planning difficult. The teams currently performing the calculations could not predict these non-flight engagements, leading to significant deviations in their calculations. Therefore, using statistical techniques and machine learning techniques, these non-flight engagements have been made predictable for the next 36 months with the fleet type and flight crew type.

A structure has been established where the model runs every month in a rolling window manner to predict the next 36 months. Additionally, an autonomous self-learning auto-ML flow has been created, allowing the model to retrain itself according to changing data, thus adapting to changes in the data over time.

This project is designed to meet the needs of crew planners in the aviation sector who want to forecast crew requirements for the long term. It is also suitable for the entire aviation sector and potentially for individuals and organizations in any sector involved in human resources planning.