休閒農場的經營須投入大量人力、物力、資金及需具備專業的經營管理能力。因此，若能有效的預測未來的營運量，就能更有效分配人力與物力，提升營運的績效。本研究透過運用類神經網路的方法，利用過去營運之數據，分析並預判未來之來客數，所得之結果，將可提供經營管理者參酌，俾達到合理的人力分配及原物料的管理甚或是營運策略之修正。休閒農場來客數預測系統建置，主要係藉由過去營運紀錄建立來客資訊，再利用適當模擬預測技術以推估未來來客數，作為後續改進農場營運措施的依據。本研究以台南市楠西區龜丹溫泉休閒體驗農園為研討案例，以其每日來園人數為輸入參數，並仔細分析過去5年（2007－2011年）的數據與來客樣式。初步結果表明，影響來客數受重要的因素包括天候（溫度、降雨量與颱風）與假日（週末與國定假日）等因子。研究結果顯示，影響龜丹休閒農園來客數的因素及權重依序分別為星期假日、國定假日、颱風及雨量。另來客數受降雨量之影響，十分顯著，但受溫度的影響則相對較不顯著，僅在低溫時（＜16°C）呈現較高的相關性。以本研究的案例而言，納入溫度、降雨量與假期等外部影響因子，使用大樣本高頻率的數據進行倒傳遞類神經網路的訓練與建模，研究結果發現以隱藏層一層及30個隱藏層節點數之神經網路架構，對休閒農場來客數的預測有極佳的效果（R2=0.932）。本研究的結果證實對於小空間尺度，使用大樣本高頻率數據，倒傳遞類神經網路仍是良好的模擬與預測的工具。

Accurate forecasting of tourism demand is essential for efficient planning by tourism-related enterprises. It helps tourism planners develop appropriate strategies; assists tourism managers in ensuring adequate capacity and infrastructure, while optimizing operational requirements. Feasibility of using one of the artificial intelligence techniques, namely, back propagation networks (BPN) to forecast tourist arrival is investigated in this study. A number of BPN models were constructed and tested using data collected from a recreation farm resort in Tainan, Taiwan. In addition, regression analysis was performed to identify critical factors influencing tourist arrival. Critical factors, as identified by order of importance, are weekends, national holidays, typhoons, rainfall, and cold weather events, respectively. Regression analysis also suggests that the first two critical factors (i.e., weekends and national holidays) together already account for approximately 90% of the influence. Furthermore, the results of this study indicate an excellent tourist arrival forecasting accuracy (R2 = 0.932) by the BPN model using one hidden layer with 30 neurons. It may be concluded that, as shown in this study, when used in an appropriate manner, the BPN model may serve as an excellent tool for tourism demand forecasting.