Airline ticket price prediction based on CNN-LSTM hybrid model
To address the need for forecasting future trends in airline ticket prices in highly competitive markets,this paper proposes a CNN-LSTM hybrid model that integrates Convolutional Neural Net-work(CNN)and Long Short-Term Memory(LSTM).In the data construction and input phase,a channel data structure is developed to represent ticket prices,emphasizing the competitive relation-ships among airlines.Furthermore,various factors influencing ticket price fluctuations are considered,leading to the creation of independent channel data structures to represent airline attributes,flight attri-butes,and date attributes.These channel data are then integrated into a multi-channel data input for-mat suitable for CNNs.In the modeling phase,a one-dimensional Convolutional Neural Network(1D-CNN)is utilized to extract features from the multi-channel input data,while the LSTM captures tem-poral dependencies within the data to predict future ticket prices for different flights on a given route.The proposed CNN-LSTM model is compared against several baseline models,and ablation experi-ments are conducted to validate the importance of the selected influencing factors.Experimental results demonstrate that the CNN-LSTM model achieves significant improvements in prediction perfor-mance.Compared with Random Forest,Support Vector Machine,standalone CNN,standalone LSTM,and Vector Autoregression model,the Mean Absolute Error is reduced by 18.74%to 57.02%,and the Mean Absolute Percentage Error is reduced by 9.31%to 22.16%.Furthermore,the ablation experiments confirm that incorporating these influencing factors enhances the model's overall performance.The findings of this study not only provide decision-making support for airlines in ticket pricing and adjustment strategies but also introduce novel methodologies and perspectives for research in airline ticket price prediction.
deep learningairline ticket predictiontime seriesconvolutional neural networklong short-term memory network
万方数据
维普
