The Intelligent Forecasting of Destination Demand for Improving Service Level Based on LSTM Model

This study aims to enhance service levels and efficiency in the tourism industry while improving tourism demand forecasting. First, it analyzes the classification of influencing factors based on tourism demand theory. Next, it develops a demand forecasting model using a deep learning framework, employing the stacked auto encoder long short-term memory (SAE-LSTM) as the benchmark structure, with an attention mechanism for optimization. This is combined with the bat algorithm-relevance vector machine (BA-RVM) model to create the BA-RVM-A-SAE-LSTM mixed model, which is then compared to traditional models and advanced mixed models. The results indicate that LSTM performs significantly better in tourism demand forecasting than other traditional models, supporting its use as the benchmark structure in this study. The mean absolute percentage error of the BA-RVM-A-SAE-LSTM mixed model is 2.63, the rank sum ratio is 0.432, the Nash-Sutcliffe efficiency coefficient is 0.992, and the goodness of fit is 0.991, all of which outperform traditional and mixed models. This study highlights the superiority of the BA-RVM-A-SAE-LSTM mixed model in forecasting accuracy and model interpretation, contributing positively to the development of the tourism industry.