來台旅遊人數與國人出國人數之時間序列預測分析

Abstract

中文摘要 隨著時代的變遷和近年來國民所得的增加，以及生活水準的提升，國內外重視休閒的人口逐漸增加，再加上政府目前實施週休二日，許多家庭外出旅遊的比例越來越高，出國觀光旅遊的人數更是快速上升，而來台的旅客人數也是只增不減，因此造就了台灣觀光產業的蓬勃發展，為國內市場帶來不少的經濟收益。為了解來台旅客人數的成長趨勢，以及國人出國旅遊狀況，本文運用在「預測分析」課程所學習到的方法進行時間序列模型的探討，並且預測未來一年來台遊客人數，以作為我國觀光產業發展的參考。 本文採用交通部觀光局觀光統計資料庫所提供的來台旅遊人數及國人出國人數進行分析，資料範圍自2010年1月至2018年12月，共計108筆月資料，並保留最後12筆進行預測。本文採用時間序列迴歸法、指數平滑法、ARIMA方法來進行模型配適，並透過殘差分析找出最佳模型。最後，再以MAE、MSE及MAPE準則來評估各模型的預測效果，並選出最佳模型。本文分析結果顯示，ARIMA方法在三個準則之下皆有最佳的表現，該模型可做為來台旅客人數與國人出國人數之預測模型。

Abstract With the changes of the times, the increase in national income in recent years, and the improvement of living standards, the number of people paying attention to leisure at home and abroad has gradually increased. In addition, the government currently implements a two-day weekly holiday, and the proportion of many families going out for tourism is increasing. The number of people going abroad for sightseeing has increased directly, and the number of tourists coming to Taiwan has only increased. Therefore, this phenomenon has caused the flourishing development of Taiwan’s tourism industry. The tourism industry will be the main force of Taiwan’s future economic development. Visitors to Taiwan and the market. So I want to use the forecasting methods learned in this lesson of forecast analysis to find the best time series model, and predict the number of tourists in Taiwan in the next year, and find out how to increase the number of tourists in Taiwan, and hope this This report can be used as a reference for the Tourism Bureau. The source of the data is the tourism statistics database of the Tourism Bureau of the Ministry of Transport. The analysis time is from January 2010 to December 2018, and the last 12 records are retained for forecasting. At the beginning of the report, confirm whether the time series graph shows a trend, or whether the variance and the average are stable. Then use time series regression, exponential smoothing, and ARIMA analysis to fit the model. Finally, use the three criteria of MAE, MSE and MAPE to evaluate which of these three methods is the best and select the best model. The analysis results show that the fit model of the ARIMA method is the best model.