A Deep Learning Approach for Predicting Rural Tourism Revenue Seasonality Using LSTM-TCN Hybrid Networks

This study employs a mixed deep learning approach combining LSTM and Temporal Convolutional Networks (TCN) to forecast seasonal tourism revenue. Unlike traditional statistical models, the proposed method incorporates socioeconomic indicators, search engine indices, and historical visitor data, enabling more accurate predictions and capturing complex nonlinear relationships. Conventional models such as ARIMA and SARIMA have long been used in economic forecasting but struggle with nonlinear patterns. In contrast, modern AI models like LSTM and CNN have proven effective for sequence-based prediction tasks. In this study, features were extracted from a clean dataset and processed through directed sequence generation before training both LSTM and TCN models. Experimental results demonstrate that the proposed hybrid model outperforms LSTM, TCN, and ARIMA in terms of MAE and RMSE, achieving an R² of 96.5%. It successfully identifies peak tourist periods, supporting government efforts in resource optimization and sustainable tourism management.