Впровадження гібридного глибокого навчання CNN-lstm для прогнозування відшкодування витрат в контексті оптимізації стратегічного управління лоукост-перевізниками в авіакомпанії PT Lion Mentari

The object of this research is the cost recovery prediction model of low-cost airlines. The challenge faced is the complexity of the airline’s financial and operational patterns, which encompass spatial and temporal interactions, making it difficult for a single prediction model to produce accurate and stable estimates. This challenge directly impacts management’s ability to develop effective strategies, including fare setting, fuel cost control, and flight route planning. To address this issue, this research implements a hybrid deep learning approach, CNN-LSTM. The CNN is used to extract spatial features from multivariate data, while the LSTM captures long-term temporal dependencies with a complex memory update mechanism. The integration of these two models allows for richer data pattern processing than a single model, resulting in more accurate predictions that align closely with actual conditions. The research interpretation indicates that the hybrid model is able to leverage the strengths of each component: CNN in extracting local features and LSTM in understanding temporal dynamics. This is reflected in the prediction results, which show a smaller deviation from the actual data compared to either CNN or LSTM alone. Based on the test results on cost recovery data for 14 periods, the three models CNN, LSTM, and CNN-LSTM hybrid showed high and stable accuracy in following the actual value pattern. From the first to the fourth period, all models produced results very close to the actual value, with an average difference below 0.02. For example, in the fourth period, the actual value of 0.88 was well predicted by CNN (0.91), LSTM (0.886), and CNN-LSTM (0.876). However, in the sixth to ninth periods, there was a slight decrease in accuracy as the actual value decreased, especially in the ninth period when the value of 0.74 was only predicted by 0.731 by CNN, 0.74 by LSTM, and 0.732 by the hybrid model. The implementation of the CNN-LSTM hybrid not only improves the accuracy and reliability of cost recovery predictions but also provides strategic value for PT Lion Mentari Airlines management, supporting more efficient and optimal decision-making to enhance the low-cost carrier’s competitiveness