E-customization of an online retailer

Official URL: https://risc01.sabanciuniv.edu/record=b2756180_(Table of contents)

Data science and machine learning algorithms enable companies to track consumer behavior from large datasets for different markets. In online retail platforms, these analyses can help to extract popular products and expose these to customers to increase the purchase probability. In this study, we used the data of an online Turkish retailer containing 841 customers and 1282 transactions to propose two frameworks to improve the recommendation system (RS) and website layout of the company to stimulate more purchases. In the rst framework, we utilized RFM (Recency, Frequency, Monetary) analysis to evaluate people’s purchasing behavior in the context of these three variables. Next, by using these features and the K-means clustering algorithm, we assigned customers to different clusters. The four segments thus built are Loyal Customers, High-potential Customers, Prosperous Customers, and Departed Customers. In the final step, relying on association rule mining, 10 products were extracted from the frequently bought itemsets of the first three clusters to offer to the customers by the recommendation system. In the second framework, we developed a mechanism that adopts the lift metric of association rule mining to change the layout of the company’s website to increase purchase probability.