Improving flight delays prediction by developing attention-based bidirectional LSTM network

Abstract

Recently, the significance of accurate aircraft delay forecasting has grown in the aviation sector, which caused multi-billion-dollar losses faced by airlines and airports and passenger loyalty losses. Due to the importance of accurate flight delay prediction for all stakeholders involved, the aviation sector seeks to develop techniques for more robust flight delay prediction. Is quickly becoming an important research issue to improve airport and airline service performance and offer customers dependable travel itineraries. Machine learning techniques have been used in a number of studies to evaluate and resolve issues with flight delay prediction. This paper proposes a framework integrated network called ‘Attention-based Bidirectional long short-term memory’ (ATT-BI-LSTM) for flight delay prediction. The Bidirectional LSTM model extracts the spatial and temporal of the flight network with weather features. The ‘Attention mechanism’ has been proposed to enable the model to discover significant and discriminating features that contribute to categorization. The first stage of the proposed framework is the ‘preprocessing of dataset’ which is performed through two steps. The first step is data transformation using MinMax scaler to reduce the variation in the data. The second step is ‘balancing the dataset’ using SMOTE technique for balancing data. The second stage is the establishment of the ATT-BI-LSTM network through the deep tuning experiments of network structure to identify the best combination of parameters and network architecture. To validate the performance of the proposed framework, a wide network of US domestic flights tested in two scenarios. In Scenario 1, the objective is to predict the delay of flight arrival and departure, by using basic flight features with the weather. In Scenario 2, the objective is to predict the delay of flight arrival, by using basic flight features with departure delays. Simulation results show that in scenario 1, the training accuracy of flights’ delay is 88% in both flight delay arrivals and departure, and the testing accuracies of flights’ delay 83% and 82% in departure and arrival respectively. On the other hand, in scenario 2, testing accuracies are 94.30% and 93.71% in the two datasets respectively. The simulation results show that the ATT-BI-LSTM model outperforms other models found in the literature. Therefore, the developed ATT-BI-LSTM framework can contribute strongly to mitigating flight delays by providing a high accuracy prediction system in real-time monitoring to airport and aviation authorities.