Historical pattern recognition with trajectory similarity for daily tourist arrivals forecasting

Forecasting daily tourist arrivals are crucial for tourism practitioners and researchers. Previous studies have shown that it is challenging to forecast the high volatility of daily tourist arrivals, especially during an emergency such as COVID-19. This study proposes a tourist arrival forecasting approach based on time series trajectory similarity (TS), which consists of five steps: (1) dividing the data into training sets, test sets, and matching sets; (2) using trajectory similarity to find the most similar historical time series within the current period; (3) data extraction, which uses the next day's data as a forecasting dataset by finding historically similar data; (4) and (5) are the evaluation of forecasting methods and results, respectively. Based on the verification before and during COVID-19, the proposed approach has achieved excellent performance in forecasting daily tourist arrivals to Siguniang Mountain.