隨著消費性電子產業的蓬勃發展，電子產品的不斷推陳出新，半導體產業近幾十年來，在現在科技發展中扮演著非常重要的角色，而動態隨機存取記憶體（Dynamic Random Access Memory，DRAM）及其DRAM記憶體模組產品，更是半導體產業中不可或缺的一環，相對的，供應鏈DRAM產業更具有舉足輕重的地位。 影響DRAM價格漲跌的原因甚為複雜，例如同業的加入與退出、製程技術的提升、市場景氣的循環、消費性電子產業的升級…等等均可能影響DRAM的價格。本研究以DRAM產業為例，希望綜合考量並彙整出影響DRAM現貨價格的因子，並使用資料探勘分析來探討DRAM現貨價格的準確度，進而提供業界更精準的採購成本預測。 若能有效的預測DRAM現貨價格走勢，對採購者而言除可增加獲利能力，更可減少投資的風險性及庫存跌價的損失，進而採取相對應的採購策略，降低公司營運險，並做為研訂公司經營策略之依據。 研究結果顯示，透過資料探勘之類神經網路分析的方法預測DRAM現貨價格的平均平方誤差 (Average Square Error) 最小，代表其預測準確度最高，可說明資料探勘的方法用於預測DRAM現貨價格走勢，是具有相當參考價值的預測模型。 ;With the development of the consumer electronics industry, semiconductor industry in recent decades plays a very important role in the field of science and technology now. DRAM (Dynamic Random Access Memory) and DRAM memory products are highly performing the outstanding segment in semiconductor industry, and DRAM industry supply chain relatively has more pivotal position. The causes of price fluctuation of DRAM products are complicated, such as new comer of DRAM manufacture or DRAM players leave, the new technology lead-in, the market cycle and the upgrade of consumer electronics products... etc. may affect the price fluctuation of DRAM products respectively. In this study, taking the DRAM industry for example, is intending to compile a comprehensive factor of DRAM spot prices and use data mining analysis to investigate the accuracy of DRAM spot prices, thereby providing a more accurate cost forecasts of procurement. If we can effectively predict DRAM spot prices, purchasers can increase profitability by reducing the risk of investment and declining the inventory loss, and then take the corresponding procurement strategy. It would further provide the basis for the high-end management strategy of the company. The results concluded that the minimum average square error (ASE) of prediction of DRAM spot prices is revealed from the neural network analysis, on behalf of its highest accuracy of prediction explain that data mining method could be used to predict the DRAM spot price movements, and it is considerable tool to predict the market price respectively.
