Prediction of Flight Time Deviation for Lithuanian Airports Using Supervised Machine Learning Model.

Area of Science:

• Aviation management
• Data science
• Machine learning applications

Background:

• Flight delays pose significant operational and economic challenges for airports.
• Accurate prediction of flight time deviations is crucial for efficient air traffic management.
• Lithuania airports' flight data has not been extensively analyzed for delay prediction.

Purpose of the Study:

• To analyze flight time deviations at Lithuania airports.
• To implement and compare supervised machine learning models for predicting flight delay intervals.
• To identify the most accurate algorithm for flight delay prediction.

Main Methods:

• Utilized seven supervised machine learning algorithms: probabilistic neural network, multilayer perceptron, decision trees, random forest, tree ensemble, gradient boosted trees, and support vector machines.
• Employed grid search for hyperparameter optimization to maximize algorithm accuracy.
• Evaluated algorithm performance using sensitivity/recall, precision, specificity, F-measure, and accuracy.
• Applied the SMOTE (Synthetic Minority Over-sampling Technique) technique for dataset balancing.

Main Results:

• Tree model classifiers, including gradient boosted trees, achieved the highest prediction accuracy.
• Gradient boosted trees were identified as the most effective algorithm for predicting flight delay intervals.
• Analysis considered both departure and arrival flights separately, incorporating weather data.

Conclusions:

• Supervised machine learning, particularly gradient boosted trees, offers a robust solution for predicting flight time deviations at Lithuania airports.
• The findings provide valuable insights for improving air traffic predictability and operational efficiency.
• Further research could explore additional features and advanced modeling techniques for enhanced delay prediction.