A hybrid machine learning-based model for predicting flight delay through aviation big data.

Area of Science:

• Artificial Intelligence
• Operations Research
• Data Science

Background:

• Flight delay prediction is a complex challenge in aviation scheduling.
• Existing machine learning methods struggle with large datasets and accuracy.
• Efficient prediction systems are vital for airlines and airports.

Purpose of the Study:

• To propose a novel, accurate, and efficient machine learning method for flight delay prediction.
• To identify key indicators influencing flight delays.
• To improve the performance of flight delay prediction systems.

Main Methods:

• Utilized ANOVA and Forward Sequential Feature Selection (FSFS) to identify influential flight delay indicators.
• Employed DBSCAN clustering to manage large flight datasets by grouping similar samples.
• Developed a novel, optimized Random Forest model weighted by the Coyote Optimization Algorithm (COA) for each cluster.

Main Results:

• Achieved a 2.49% increase in accuracy and a 39.17% improvement in processing speed through clustering.
• The COA-optimized Random Forest model showed a 5.3% accuracy increase over conventional models.
• The proposed method reached an average accuracy of 97.2%, a 4.7% improvement over prior research.

Conclusions:

• The integrated approach of feature selection, DBSCAN clustering, and COA-optimized Random Forest significantly enhances flight delay prediction.
• The method offers a scalable and accurate solution for real-world aviation challenges.
• This research provides a robust framework for improving operational efficiency in air travel.