Predicting Purchase Interest of Online Shoppers Using Boosting Algorithms

Abstract

Data driven marketing is becoming more and more vital for businesses day-by-day. Understanding customer behavior has the potential to decrease marketing costs as well as increase sales both in conventional marketing and online marketing. Since online users can access information faster, prices have become more competitive and customer behavior analysis has become more important. The purpose of this study is to predict the purchase interest of the users in an e-commerce web page by using the user session data such as pageview, duration etc. To this aim we used clickstream data for an e-commerce web page which is publicly available. Since only 16.5 percent of the sessions are completed with purchase in the dataset, increasing true positive rates rather than accuracy is more important. To this aim, we have explored the performance of boosting algorithms on the dataset and compared to those of state-of-the-art methods that were previously applied on the same dataset. Results show that boosting algorithms have better performance for identification of the sessions that end with a purchase.