資料探勘為基礎之零售業銷售預測模式-以連鎖超商鮮食商品為例

Constructing a Retailing Sales Forecasting Model Base on Data Mining Framework: An Empirical Study on Fresh Food in Chain Store

10.6843/NTHU.2010.00643

歐宗殷

連鎖超商 ； 鮮食商品 ； 銷售預測 ； 灰關聯分析 ； 統計時間序列 ； 類神經網路 ； 田口方法 ； Convenience Store ； Fresh Food ； Sales Forecasting ； Gray Relation Analysis ； Statistical Time Series ； Neural Network ； Taguchi Method

清華大學工業工程與工程管理學系學位論文

2010年

博士

陳飛龍

繁體中文

在連鎖超商的經營管理中，管理者每日都必須預估未來的需求，並依此訂購適量的商品，過多的存貨將產生多餘的浪費，而存貨不足又可能因而失去潛在的商機，準確的預測模式將可有效地增加獲利、降低成本並有利於消費者滿意度的提升。 面對商品銷售的預測問題，連鎖超商業者莫不期望能有更好的決策支援系統來協助此項工作，以便能在快速變動的商業環境之下洞燭機先，假設管理門市的經理人能預先掌握並得知消費者對於商品的需求趨勢以及影響因子，則顧客的需求將有效地被滿足，同時報廢的商品也將大幅減少。 本研究以資料探勘的方法論建構銷售預測模式，運用灰關聯分析（Grey Relation Analysis；GRA）的特性，將影響銷售量的重要因子篩選出來，而銷售預測模式的建構則是以統計時間序列中MA、ARIMA與GARCH，以及與類神經網路中BPN、MFLN和ELM為主，並以實際的銷售資料進行預測模式的比較，預測誤差是以MSE、MAD及THEIL三項指標進行評估，所得結果顯示，本研究建構的GELM預測模式其預測能力不僅比時間序列等模式為佳，與GBPN和GMFLN相較之下，不但預測能力較佳，且訓練時間更大幅縮短。 由於活化函數與隱藏層層數將影響GELM預測模式準確度及學習速度，本研究運用田口方法中的直交表進行實驗設計，經實驗確認後證明，當隱藏層為25層並採用hardlim為活化函數時，可使GELM模式的預測誤差最小化。

Due to the strong competition that exists today, most retailers are in a continuous effort for increasing profits and reducing their cost. An accurate sales forecasting system is an efficient way to achieve the aforementioned goals and lead to improve the customers’ satisfaction, reduce destruction of products, increase sales revenue and make production plan efficiently. While manage the convenience store, the supervisor should estimate the daily demand of the future and place an order to purchase the commodities. If the managers can estimate the probable sales quantity in the next period, the demand could be satisfied and the cost of spoiled fresh foods would substantially be reduced. Besides a good forecasting model leads to improve the customers’ satisfaction, reduce destruction of fresh food, increase sales revenue and make production plan efficiently. This study constructs a retailing sales forecasting model by data mining framework. Firstly, it applies GRA to realize the relationship between two sets of time series data in relational space then sieves out the more influential factors from raw data and transforms them as the input data for developing the forecasting model. Secondly, this research applies time series forecasting model includes MA, ARIMA, GARCH and neural network forecasting model includes BPN, MFLN and ELM. The proposed system evaluated the real sales data in the retail industry. The experimental results demonstrate that our proposed system which integrates GRA and ELM based on robust experiments design with Taguchi method outperforms than other sales forecasting methods based on time series and neural networks methodology.

1. 張文娟(2016)。以MARS模式探討時尚產業虛實通路之銷售影響因素。淡江大學管理科學學系碩士班學位論文。2016。1-53。 
2. 王品媁(2016)。智慧型日交易量建模與預測於台灣股票與期貨市場之研究。交通大學資訊管理研究所學位論文。2016。1-65。