- 期刊
中油Pay會員加油行為分析與流失預測
CPC Pay Membership Analysis and Churn Prediction
摘要
處在網路數位經濟時代，消費者的購物管道、支付方式、瀏覽習慣與決策行為皆與傳統顯著有別。隨著數位工具和電子支付越來越普及，以及大數據分析、人工智慧演算法的進化等現象的加成下，企業可透過數據分析了解消費者，使得消費者的組成、習慣與需求等輪廓逐漸清晰，進而可區分高價值客群與低價值客群等，並進行客戶流失預測。為不同客群規劃行銷策略與活動，不僅能夠降低行銷成本、節省廣告預算，更能促進商機、增加營收。本研究主要為客戶交易資料之延伸應用，利用大數據技術分析市場現況及客群組成，以鎖定具備高消費潛力的顧客。本研究利用中油Pay會員交易資料，使用RFM（Recency, Frequency, Monetary）模型結合K-Means演算法進行中油Pay會員加油行為分析，並取得客戶流失定義，進而以多種機器與集成學習，預測未來中油Pay客戶是否將流失。研究結果顯示，準確度可達81.65%，召回率近8成。模型對於區分中油Pay客戶價值高低，以及預測中油Pay客戶未來是否流失有良好效果，有利於台灣中油公司決策單位規劃個人化行銷策略以投其所好、精進服務品質，進而避免客戶流失。
並列摘要
In the era of digital economy, consumer shopping channels, payment methods, browsing habits, and decision-making behaviors have significantly deviated from traditional approaches. With the widespread adoption of digital tools and electronic payments, coupled with advancements in big data analytics and artificial intelligence algorithms, businesses can leverage data analysis to understand consumers better. This enables the gradual delineation of consumer profiles, habits, and demands, allowing differentiation between high-value and low-value customers, as well as customer churn prediction. By tailoring marketing strategies and activities to different customer segments, companies can not only reduce marketing costs and save advertising budgets but also foster business opportunities and increase revenue. This study focuses on the extended application of customer transaction data, employing big data techniques to analyze market conditions and customer segments, with the aim of targeting customers with high consumption potential. Using transaction data from CPC Pay, the study combines the RFM (Recency, Frequency, Monetary) model with the K-Means algorithm to analyze the refueling behavior of CPC Pay members. By defining customer churn and employing various machine learning and ensemble learning techniques, the study predicts whether CPC Pay customers will churn in the future, achieving an accuracy rate of 81.65% and effectively identifying churned customers among nearly 80% of CPC Pay users. The results demonstrate the effectiveness of this study in differentiating the value of CPC Pay customers and predicting their future churn, thereby aiding decision-making units of CPC in planning personalized marketing strategies, improving service quality, and preventing customer churn.