Hybrid LSTM–TCN Architecture for Seasonal Income Forecasting in Rural Tourism Using Macroeconomic and Tourism Indicators

Abstract

This study proposes a Hybrid RTLSTM–TCN deep learning architecture for forecasting seasonal income in rural tourism using integrated macroeconomic and tourism indicators. The RTLSTM component captures long-term sequential dependencies, while the TCN block models short-term temporal variations through dilated causal convolutions. The model was evaluated against benchmark approaches including ARIMA, KELM, MSS-KELM, B-SAKE, RNN, BiLSTM-TN, and SAE-LSTM. Empirical results on multi-year tourism datasets demonstrate that the proposed RTLSTM–TCN achieves the lowest RMSE (0.18) and MAE (0.09) with the highest R² (0.85), outperforming existing machine learning and deep learning baselines. This approach improves forecasting robustness under seasonal and macroeconomic volatility, offering a decision-support tool for tourism policy planning and economic sustainability.