Repurchase Prediction Of New Customers Based On Browsing History And Customer Clustering

With the gradual increase in customer acquisition costs,how to guide new customers to convert into repeat customers has become a problem faced by e-commerce platform merchants.Since it is not necessary to consume additional marketing resources for new customers with a high likelihood of repurchase,and it is too difficult to retrieve new customers with a low likelihood of repurchase,it is essential to identify hesitant new customers who are hesitant to repurchase and develop marketing strategies for these potential repeat customers in order to reduce costs and maximize benefits.This paper uses the browsing history of new customers for six months on the whole ecommerce platform and their personal information such as gender and age.After data preprocessing procedures such as data cleaning and feature engineering,XGBoost,Light GBM and Cat Boost are used to build classification models to estimate the repurchase probability of new customers.After that,new customers are divided into groups according to the repurchase probability,and new customers with repurchase probability between 46% and54% are classified as hesitant new customers who are more worthy of investment in marketing resources under the premise that the total number of hesitant new customers is moderate.After clustering new customers,measures are taken to improve the capability of the model for hesitant new customers.When imputing the missing values of the category variables,treating the missing values as a special category can lead to an average improvement of 3.34%on hesitant new customers compared with the commonly used mode imputation.Meanwhile,the stacking approach is improved by weighting the extracted features according to the base learner’s differentiation of hesitant new customers,and the model after stacking has a 2.30% increase in effectiveness on this group compared to using only a single model.Finally,by analyzing the factors influencing new customers’ willingness to repurchase through feature importance,it is found that the number of existing repeat customers in the store is significantly more important than other features,and other important factors could be roughly divided into two aspects: the number of existing customers in the store and the market situation of sold products.These results can help stores on e-commerce platform for the further development.