蘭花為臺灣具高經濟價值之重要農作物，外銷實力居世界前三大蘭花出口國之列。然而，由於其生產週期長達數月甚至一年，產銷過程長期面臨市場需求波動劇烈與預測不確定性高等挑戰。多數業者仍以經驗法或歷年同期數據作為銷售預估依據，難以及時因應節慶需求、經濟景氣變動或突發事件（如疫情）所造成的市場劇烈變化，導致庫存累積與營運成本上升。
為此，本研究旨在建立一套具可行性與解釋力之量化預測模型架構，以協助業者提升生產規劃與資源配置效率。本研究以自迴歸模型（Autoregressive Model, AR）為核心，用以捕捉蘭花歷史銷售資料中的時間序列結構與自我相關特性。同時，為了提高預測準確度與模型解釋力，導入遞迴特徵消除交叉驗證法（Recursive Feature Elimination with Cross-Validation, RFECV）進行特徵選取，從花色、節慶與價格等多項變數中篩選出對需求變化最具影響力之關鍵特徵變數。此外，採用滑動視窗法進行動態模型訓練與驗證，使模型能適應資料隨時間變動的特性。
綜上所述，本研究期望藉由比較納入特徵變數前後模型的預測表現，建構一套具結構清晰、統計基礎穩健且可實務應用的蘭花銷售量預測系統，以協助業者在高度不確定的市場環境中，做出更精準且數據導向之決策。

Taiwan is among the world's top three orchid exporters, yet the long production cycle and volatile market demand make sales forecasting highly uncertain. Most growers still rely on expert experience and same-period historical figures, which leads to overstocking and elevated operating costs. This study develops a quantitative and interpretable forecasting framework for monthly orchid sales. An autoregressive (AR) model serves as the baseline; an ARIMA model captures richer time-series structure; and an ARX model augments AR with exogenous features selected through Recursive Feature Elimination with Cross-Validation (RFECV). A sliding-window scheme is used for dynamic training and six-step-ahead out-of-sample validation, and statistical errors are translated into an asymmetric operating-cost matrix. Across 470 comparable samples, ARIMA achieved the lowest proportional error (MAPE), whereas ARX achieved the lowest absolute error (MAE and RMSE) and the largest cost reduction relative to AR. Product attributes such as flower color contributed more than festival effects. The results show that the statistically most accurate model is not necessarily the most cost-effective, and that model selection should depend on the decision objective and the available data conditions.

Ahaggach, H., Abrouk, L., and Lebon, E. (2024). Systematic mapping study of sales fore-casting: Methods, trends, and future directions. Forecasting, 6(3):502–532.

Arunraj, N. S., Ahrens, D., and Fernandes, M. (2016). Application of sarimax model to forecast daily sales in food retail industry. International Journal of Operations Research and Information Systems (IJORIS), 7(2):1–21.

Badal, P. S., Kamalvanshi, V., Goyal, A., Kumar, P., and Mondal, B. (2022). Forecasting potato prices: application of arima model. Economic Affairs, (4):491–496.

Eiglsperger, J., Haselbeck, F., Stiele, V., Serrano, C. G., Lim-Trinh, K., Menrad, K., Han-nus, T., and Grimm, D. G. (2024). Forecasting seasonally fluctuating sales of perishable products in the horticultural industry. Expert Systems with Applications, 249:123438.

Hossain, M. M. and Abdulla, F. (2016). Forecasting potato production in bangladesh by arima model. J Adv Stat, 1(4):191–198.

Huber, J. and Stuckenschmidt, H. (2020). Daily retail demand forecasting using machine learning with emphasis on calendric special days. International Journal of Forecasting, 36(4):1420–1438.

Jahin, M. A., Shahriar, A., and Amin, M. A. (2025). Mcdfn: supply chain demand forecasting via an explainable multi-channel data fusion network model. Evolutionary Intelligence, 18(3):66.

Jenčová, S., Vašaničová, P., Košíková, M., and Miškufová, M. (2025). A time series approach to forecasting financial indicators in the wholesale and retail trade. World, 6(1):5.

Kittichotsatsawat, Y., Boonprasope, A., Rauch, E., Tippayawong, N., and Tippayawong, K. Y. (2023). Forecasting arabica coffee yields by auto-regressive integrated moving av-erage and machine learning approaches. AIMS Agriculture & Food, 8(4).

Li, J., He, D., and Lu, X. (2024). Sales forecasting and classification model and pricing research of agricultural perishables based on stl-arima. Highlights in Business, Economics and Management, 33:281–288.

Luis-Rojas, S., Ramírez-Valverde, B., Díaz-Bautista, M., Pizano-Calderón, J., and Ro-dríguez López, C. (2020). Vanilla (vanilla planifolia) production in mexico: analysis and forecast.

MacLachlan, M. J., Adjemian, M. K., Etienne, X., Sweitzer, M., Volpe Iii, R., and Zeng, W. (2025). Adaptive food price forecasting improves public information in times of rapid economic change. Nature Communications, 16(1):6282.

Observatory of Economic Complexity (2023). Orchids (hs: 060313). https://oec.world/en/profile/hs/orchids?selector1013id=2023&selector1699id=usdOption&selector1777id=valueOption&selector2367id=Year.Retrieved August 7, 2025, from the Observatory of Economic Complexity website.

Sandeep, K., Sharma, S., Sharma, A., and Lohia, m. (2025). Forecasting of rice production in india using linear time series models. 17:19.

Sanjith Bharatharajan Nair, R. E. (2023). An autoregressive integrated moving average model approach in agriculture: A case study. UGC and HEI Recognised Journal, 10:703–718.

Sujjaviriyasup, T. and Pitiruek, K. (2013). Hybrid arima-support vector machine model for agricultural production planning. Applied Mathematical Sciences, 7(57):2833–2840.

TogaCloud 資訊中心 (n.d.). 台灣蘭花出口統計資料查詢. https://www.togacloud.org.tw/InfoCenter?id=11653. Retrieved Month Day, Year, from TogaCloud資訊中心 website.

Yoo, T.-W. and Oh, I.-S. (2020). Time series forecasting of agricultural products’sales volumes based on seasonal long short-term memory. Applied sciences, 10(22):8169.

臺南市政府主計處 (2023). 臺南市花卉生產及市場銷售分析報告 [pdf]. Available from https://share.google/cAVWWIf2r9LSe5DBc. Retrieved Month Day, Year, from 臺南市政府主計處 website.