運用資料探勘與Spark分散式架構之智慧零售分析

Abstract

近年來傳統販賣機開始智慧化，逐漸演變成能接受多元化支付以及販售的系統。智慧販賣機可透過物聯網取得銷售、存貨和客戶偏好等數據，即時做出反應與調整，從而讓營運成本下降並提升銷售。因此，建立良好的預測模型則是智慧零售的重要議題。 本研究分析智慧零售資料以建立銷售預測模型及商品推薦模型。銷售預測包含一般產品銷量預測以及針對短周期或限量試用商品的銷量預測。結合分群與機器學習方法來建立預測模型。商品推薦則是利用銷售量資訊，評估推薦各地點智慧販賣機機台適合販賣的商品，並比較分群與未分群的推薦結果。本研究的資料來源是智慧零售販賣機，七個月的銷售資料。研究分析結果期望能達到商品銷售效益的最大化。

Traditional vending machines have recently been changed to intelligent vending machines, which can use IoT (Internet of Things) and various payment systems to lower the cost and react immediately by analyzing the collected sales and customer preference data. Therefore, it is important for retail business to deploy appropriate intelligent sales prediction models that can help business adopt further decision strategy. This research analyzes retail sales data to build sales prediction and product recommendation models for intelligent vending machines. The sales predictions contain the normal product sales prediction and the free trial product sales prediction. The proposed approach combines clustering and machine learning techniques to construct the prediction models. The product recommendation model analyzes sales information to recommend suitable products for intelligent vending machines. This research analyzes the sales data collected from intelligent vending machines. The prediction results of different models are compared. The cluster-based and non-cluster-based models are also compared. The research result will contribute to maximize the benefit of product sales.