利用資料探勘技術設計花卉運銷專家系統

Abstract

本研究利用資料探勘技術設計一個花卉運銷專家系統。在此專家系統中，共包含三個主要的功能。在第一個功能方面，利用資料探勘中決策樹之C4.5演算法找出玫瑰花殘貨法則，這些法則可以用來將花卉資料倉儲中的交易樣本歸類，以判斷其殘貨狀況；之後將這些法則彙總整理，存入知識庫中，以建立專家系統，此系統可以提供各供應單位以網際網路為基礎的專家級決策支援，以避免將來花卉殘貨的發生。在第二個功能方面，結合資料探勘中之階層式集群分析和非階層式集群分析演算法以找出優良花卉供應商，所使用的指標為殘貨比例和平均拍賣價格，優良供應商是指在同一花卉等級之下殘貨比例低且平均拍賣價格高的供應單位，此優良供應單位可以當做標竿，作為其他供應單位的模範。在第三個功能方面，利用變異數分析技術探討數種玫瑰花之等級和價格間的相關性，研究結果顯示玫瑰花等級越高，其拍賣價格也越高，這將鼓勵各花農供應較高等級的花卉，以獲取較高的拍賣價格。結合花卉運銷專家系統中三個功能所產生的花卉知識，將使得花卉市場有更佳的效率和效能。

Data-mining techniques have been applied to designing a flower marketing expert system. In this expert system, there are three major functions. First rules for classifying transactional samples from a flower data warehouse are generated using an entropy-based decision tree algorithm named C4.5. Then the rules are collected in the implemented expert system for avoiding the occurrence of overstocked flowers in the future trading. Second hierarchical and nonhierarchical clustering algorithms are combined together for finding excellent flower suppliers. Two chosen criteria for excellent suppliers are lower overstocked flowers and higher average price for the same graded flower in the market since both imply they offer the more qualified flower than others. Third techniques in analysis of variance is used for exploring the correlation between the prices and the grades of various roses. Such a positive correlation research result offers a price incentive for supplying higher graded roses in the market. Also the three functions of the underlying expert system are developed toward the direction of making the flower market more efficient and effective.