花卉產業為國內農業發展之重點指標,農政單位也因此輔導成立花卉批發市場,以期提供產銷雙方更透明的交易平台。目前國內近九成花卉交易多以電腦化拍賣進行,拍賣成交價格對於市場各方而言都是非常重要的資訊,直接影響產業各方之經營決策制定;因此,花卉成交價格之預測自然為各方決策者所重視。本研究以隱藏馬可夫模型(Hidden Markov Model, HMM)為預測方法,透過歷史拍賣價格訓練模型。本研究結果顯示,以HMM預測模型針對測試資料進行預測,平均絕對誤差百分比(MAPE)為7.64%,可視為高度準確性預測,而且在連續拍賣與商品間品質落差較小時,對拍賣成交價格預測準確度能達最好之成效。本研究同時與移動平均法、指數平滑法及ARIMA模型之預測結果比較,實驗結果顯示,在本研究的案例中,HMM於花卉拍賣價格預測能力優於其他三個方法。
Floriculture industry is one of the most important drivers of Taiwan’s agricultural development. With government’s supports, public flower transaction marketplaces have been established, and nowadays over 90% of flower transactions are carried out by computerized auctions at the marketplaces. The price information directly affects all stakeholders’ operation strategies, and thus the forecasting of the auction price is a critical concern to all decision-makers in this industry. This study employees the Hidden Markov Model (HMM) to forecast the auction prices of flowers. The model is trained by historical auction and tested by unseen data. The computational results show that the mean absolute percentage error (MAPE) with the forecasting is 7.64%, which can be considered as a highly accurate forecasting. The performance of the HMM model is also compared with that of a Moving Average (MA) model , an Experiential Smoothing (ES) model and an ARIMA model , and the results demonstrate that the HMM model outperforms the others.