從預測為流失的顧客中找出特徵以識別出較易挽留的顧客群集;Find Out the Feature to Identify the Cluster in Churn Customer that is More Possible to be Retained

NCUIR > School of Management at National Central University > Graduate Institute of Business Administration > Electronic Thesis & Dissertation > Item 987654321/94846

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/94846

Title:	從預測為流失的顧客中找出特徵以識別出較易挽留的顧客群集;Find Out the Feature to Identify the Cluster in Churn Customer that is More Possible to be Retained
Authors:	陳俊亨;Chen, Chun-Heng
Contributors:	企業管理學系
Keywords:	顧客流失;顧客挽留;SMOTE;K-MEANS;Customer Churn;Customer Retention;SMOTE;K-MEANS
Date:	2024-07-06
Issue Date:	2024-10-09 15:33:29 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著人們越來越重視的身體的健康，社會對於保健食品的需求也愈發強烈，需求也帶來了更激烈的產業競爭，這意味著消費者有著更多元的商家選擇。顧客流失指的是顧客在特定的期間不再向企業購買產品，而對企業來說，挽留顧客是一個非常重要的環節，因為它有著相較於獲取新顧客更低的成本，所需付出的成本差異達20倍之多，而如何找到較易挽留的顧客更是一大課題。在保健食品銷售的眾多通路中，以電話行銷這個通路最為常見，傳統上電話行銷人員需要針對名單上的顧客大量撥打電話來促使成交，然而盲目的撥打是一種不必要的人力成本耗費。相較會持續保持忠誠的顧客，企業應針對未來會流失的顧客進行挽留，但當中又有些顧客是堅定的流失而無法挽留，也就是說，企業應針對那些未來會流失，但卻較可能被挽留的顧客群體展開行動。本研究揭示了從真實的公司交易、通話資料解析出可用特徵，到增加特徵、篩選特徵、決定採用期數的過程，結合了預測模型和六種SMOTE方法找出流失顧客，其中L2正則化的羅吉斯迴歸搭配ADASYN在不同的超參數k之下，Precision平均可達82.224%，意即我們預測為流失的顧客，達到八成以上的比例確實為流失。後再以K-MEANS方法對預測為流失的顧客進行分群，並找出特徵去識別出當中較易被挽留的顧客群集。 ;As people pay more attention to their health, society′s demand for health supplements has become stronger. The demand has also brought about more intense industrial competition, which means that customers have more diverse merchant choices. Customer churn refers to the customer no longer purchasing products from the company in a specific period. For companies, retaining customers is very important because it has a lower cost than acquiring new customers. The cost difference is as much as 20 times. Among the many channels for selling health supplements, telemarketing is the most common. Traditionally, telemarketers need to make a large number of calls to customers on the list to promote purchase. However, blind calling is an unnecessary waste of labor costs. Compared with customers who will continue to remain loyal, companies should target customers who will churn in the future, but some of them are determined to churn and cannot be retained. In other words, companies should target and take action to the customers who will churn in the future but are more likely to be retained. This study reveals how to parse available features from real company transactions and call data, to the process of adding features, filtering features, and deciding the number of periods to adopt. It combines predictive models and six SMOTE methods to find churn customers, among which L2 regularization using Logistic regression and ADASYN under different k, the average precision can reach 82.224%, which means that more than 80% of the customers predicted to be churn are actually churn. Then, cluster the customers predicted to be churn through K-MEANS method, and find out the feature to identify the customer cluster that is more possible to be retained.
Appears in Collections:	[Graduate Institute of Business Administration] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	41	View/Open

社群 sharing

Loading...