Flight Delay Regression Prediction Model Based on Att-Conv-LSTM.

Area of Science:

• Artificial Intelligence
• Data Science
• Transportation Systems

Background:

• Accurate flight delay prediction is crucial for mitigating large-scale disruptions.
• Current regression models often overlook spatial data, relying solely on time series analysis.
• This limitation hinders the precision of flight delay forecasting.

Purpose of the Study:

• To propose an improved flight delay prediction method by integrating spatio-temporal features.
• To enhance prediction accuracy by addressing the limitations of single time series networks.
• To evaluate the effectiveness of an attention mechanism in improving model performance.

Main Methods:

• A novel flight delay prediction method based on Att-Conv-LSTM is introduced.
• Long short-term memory (LSTM) networks extract temporal features.
• Convolutional neural networks (CNNs) extract spatial features.
• An attention mechanism module is integrated to optimize network efficiency.

Main Results:

• The Conv-LSTM model demonstrated an 11.41% reduction in prediction error compared to single LSTM.
• The Att-Conv-LSTM model achieved a further 10.83% reduction in prediction error compared to Conv-LSTM.
• These results confirm the benefits of incorporating spatio-temporal characteristics.

Conclusions:

• Integrating spatio-temporal data significantly enhances flight delay prediction accuracy.
• The proposed Att-Conv-LSTM model offers a superior approach for flight delay forecasting.
• Attention mechanisms effectively improve the performance of deep learning models in this domain.