Imputation recovery tourism demand forecasting
针对旅游需求数据缺失和外部冲击导致的历史数据不可靠问题，提出RTD框架，利用冲击前趋势重建序列并估计恢复，结合深度学习和时间序列分解提升预测准确性，优于传统模型。
Tourism demand forecasting is vital for planning yet faces challenges like data misalignment and external shocks. Data misalignment—due to missing values, irregular reporting, and inconsistent frequencies—undermines data quality. Shocks such as pandemics or natural disasters disrupt patterns, reducing the reliability of historical data. Traditional models (e.g., ARIMA, ETS) assume clean, regular data, limiting their effectiveness in such contexts. While preprocessing (e.g., imputation) helps, it can cause information loss. This study proposes the RTD (Recovery Tourism Demand Forecast with Imputation) framework, which reconstructs disrupted series using pre-shock trends, then estimates recovery to adjust forecasts. Combining deep learning and time series decomposition, RTD minimizes data loss and improves accuracy. Results show RTD outperforms conventional models, aiding recovery-focused tourism planning. • Define the data misalignment in tourism demand data. • Providing the solution RTD for the tourism demand forecasting under disruption. • Case studies across 20 destinations with outperforming performance.