- 學位論文
演化式支援向量迴歸於旅遊需求之預測
An Evolutionary Support Vector Regression in Tourism Demand Forecasting
摘要
由於觀光產業具備易消逝性（Perishability）的經濟特徵，也就是說，其產品具有無法儲存（如旅館空房、航空班機座位等），產出後即消逝之特性，因此其供給無法隨著市場的需求或變化而任意加以調整，而必須在供需方面加以配合管理，由此可知旅遊業之需求預測將顯的格外重要。所以若能對旅遊需求預測得當，則將可預先得知未來旅遊需求的相關統計資料，這些預測資料將會對計畫旅遊者或旅遊政策制訂者，於其旅遊決策的形成過程中帶來重要的影響。 支援向量機（Support Vector Machine, SVM）的發展最先被運用在模式識別領域，然而隨著ε-不敏感損失函數（ε-insensitive loss function）的導入，支援向量機已經被擴展到解決非線性迴歸估計的問題上，此類技術稱為支援向量迴歸（support vector regression, SVR）。本研究將運用支援向量迴歸技術建構旅遊需求之預測模型。研究中將提出一種名為GA-SVR的新模型，該模型先運用實數值遺傳演算法以找出支援向量迴歸的最佳化參數，並藉這些最佳化參數值建構支援向量迴歸模型，以預測旅遊需求，並將其預測結果和類神經網路模型（neural networks）進行比較，以證明GA-SVR模型確實擁有優良的預測能力。此外為了驗證參數對支援向量迴歸模型的重要性並獲知支援向量迴歸模型的一些特性，研究中使用了敏感度分析技術（sensitivity analysis），該分析中也證實了，不當的選取參數將使模型容易陷於過度擬合（over-fitting）或不足擬合（under-fitting）的危機中。
並列摘要
The tourism industry, which benefits the transportation, accommodation, catering, entertainment and retailing sectors, has been blooming in the past few decades. The 20th century witnessed a steady increase in tourism all over the world. Each country wants to know its international visitors and tourism receipts in order to choose an appropriate strategy for its economic well-being. Hence, a reliable forecast is needed and plays a major role in tourism planning. Support vector machine(SVM) is first applied in pattern recognition problem, however, introduction of ε- insensitive loss function by Vapnik, SVM has been developed in solving non-linear regression estimation problem, such new techniques called support vector regression(SVR). This study will apply support vector regression technique to construct the prediction model of tourism demand. In order to construct effective SVR model, we have to set SVR’s parameters carefully. This study proposes a new model called GA-SVR that searching for SVR’s optimal parameters through applies real-valued genetic algorithms, and uses the optimal parameters to construct SVR model. The conclusion in the forecasting the tourism demand study. SVR model shows its reliabilities and as a good prediction technique. Its generalization performance is even more accurate than neural networks. Moreover, in order to test the importance and understand the features of SVR model, this study implements sensitivity analysis technique, the analysis demonstrates that incorrectly selected parameters will lead the model’s results in the risk of over-fitting, or under-fitting.