LRFS: online shoppers’ behavior based efficient customer segmentation model

Abstract

The popularity of online shopping has grown significantly across the globe in recent years. This research proposes a customer segmentation model LRFS, an extended version of LRF model, built specifically for online shopping, using a dataset that includes some features taken from Google Analytics. It introduces component S, which measures the Staying Rate across the Revenue spent by the customers on a particular website, to get a better insight into the customer base. Three wellknown clustering methods K-Means, K-Medoids, and DBSCAN algorithms were incorporated along with the proposed model. For each of these algorithms, the dataset was compressed separately using three different dimensionality techniques such as PCA, t-SNE, and Autoencoder to figure out the combinations that could work well for the used dataset. A comparative analysis has also been conducted among LR, LF, LRF, and the proposed LRFS model using K-Means clustering. LRFS has outperformed the other three models in terms of better cluster assignment of the customers. For customer analysis, a combined Customer Classification and Customer Relationship Matrix was used to determine the clustered groups according to their characteristics. The combination of K-Median and t-SNE was chosen for the final combined matrix since it had the highest number of most distinct clusters with all traits of customer groups. Finally, some test cases as well as related use case scenarios have been described and visualized using LRFS along with K-Means and PCA.