最後，分群模型歸類為五群，挖掘出兩大高價值忠誠顧客族群，其他群族分別為高消耗顧客、不確定型新顧客、不確定型流失顧客等。模型特徵評估水準上，預期購買次數為R-squared為0.850、預期購買金額為 R-squared為0.877，預測特徵上有著良好的表現。將預測模型結合分群模型，分析不同群體間的特徵表現，為企業制定客製化行銷策略，提升企業的營運績效並降低整體成本，維護長期與顧客之互動關係，作為企業參考的依據。 ;In today’s turbulent food e-commerce retail industry, businesses need to remain just as competitive while simultaneously establish long-term interaction with customers to increase profits. For many enterprises, customer segmentation enables them to effectively understand customer characteristics and allocate appropriate resources between different customers. Enterprises must also simultaneously respond to changes in consumer behavior and discover insights by applying probabilistic models.
There are two major objectives in this research. First is to develop LRFM model with data mining approach for customer segmentation. Second is to apply BG/NBD and Gamma/Gamma models to predict customer purchase behavior. Both models are generally used in non-contractual market environments to measure customers’ long-term profitability. Results indicate that customer profiles can be classified into five groups, including high-cost consuming group, uncertain new customer group, uncertain lost customer group, and two groups of loyal high value customers. Regarding the accuracy of prediction models, the R-squared of the BG/NBD model for purchase frequency is 0.850, while the R-squared of the Gamma/Gamma model for purchase amount is 0.877. The results of the prediction models are integrated with segmentation model to analyze feature performance between different groups. The proposed models can help enterprise develop customized marketing strategies to bolster operational performance, while lower overall costs and maintain long-term interaction with customers.
