倒傳遞網路與決策樹於備用零件需求預測模式分類之研究

備用零件不是屬於銷售給消費者的中間產品或最終產品，而是在生產活動中維持生產設備順利運轉不可或缺的要素。備用零件的功能是確保生產設備保持在運行狀態，對於其需求量能準確地分析預測，是備用零件庫存管理的關鍵要素。大部分的備用零件單價高昂，過高的備用零件庫存水準可能會導致企業資金週轉問題，有效控管備用零件的存貨水準是企業的重要課題。
本研究收集專門生產工業原料的某製造業公司的二年備用零件歷史消耗資料，運用倒傳遞網路及決策樹C5.0演算法分別建構備用零件最適需求預測方法之分類模式。同時，實證以不同的週期數建構移動平均法與移動拔靴法的需求預測模式，並探討其對需求預測結果的影響程度。實證結果顯示，倒傳遞網路分類模式及決策樹C5.0演算法分類模式的準確度皆在七成以上且極為相近，顯示前述二者皆可用在分類各品項的最適需求預測方法上。研究同時發現不管是移動平均法或者移動拔靴法都是以週期數為十二期的模式績效最佳。

Spare parts are not intermediate products or final products that are sold to consumers, spare parts are an essential part of maintaining smooth operation of production equipment during production activities. The function of the spare part is to ensure that the production equipment is kept in operation, and the analysis of the demand forecast is a key element of the inventory management of the spare parts. Most spare parts are expensive that too high spare parts inventory levels may lead to the company's capital turnover problem. Therefore, effectively controlling the inventory level of spare parts is an important issue for enterprises.

This study collects two-year historical data of spare parts of a manufacturing company specializing in the production of industrial raw materials. We propose classification model of Back-Propagation Neural Network and Decision Tree in demand forecasting model of spare parts. We analyze the influence of different period database on both Moving Average forecasting model and Moving Bootstrap model. The results show that the accuracy of both Back-Propagation Neural Network classification model and Decision Tree classification model are more than 70%, showing that both two classification models can be used to classify the optimal demand forecasting model for spare parts. This study also found that both the Moving Average forecasting model and Moving Bootstrap model performed best in the model with a period of twelve.