Publication: A comparative study of repeat buyer prediction: Kaggle acquired value shopper case study
0
0
Issued Date
2019
Resource Type
File Type
application/pdf
Other identifier(s)
2-s2.0-85066955412
Rights Holder(s)
Scopus
Bibliographic Citation
ACM International Conference Proceeding Series. Vol Part F148384, (2019), p.306-310
Suggested Citation
Charanasomboon T., Viyanon W. A comparative study of repeat buyer prediction: Kaggle acquired value shopper case study. ACM International Conference Proceeding Series. Vol Part F148384, (2019), p.306-310. doi:10.1145/3322645.3322681 Retrieved from: https://hdl.handle.net/20.500.14740/5588
Author(s)
Abstract
Many consumer brands try their best to offer promotions that attract new customers with that hope the customer will remain loyal to the brand and come back to buy more. However, only a fraction of customers who use these promotions actually remained loyal after the promotion period. This type of person typically bought products because they had a promotion (one-time deal hunter) As a result, promotional campaigns have become less effective. Thus, the prediction model is now a popular tool to make predictions about customers who remained loyal after the promotional period to make it more targeted and maximize its effectiveness. In this paper, a solution was proposed on repeat buyer prediction in order to identify buyers with the potential to come back to buy more products. This solution would help companies reduce their budgets in terms of distributing promotional offers to customers, not just a one-time deal hunter. In this study, the dataset used was called "Acquire Value Shopper Challenge", which focused on customers transactions and used a promotional offering information on the Kaggle website. The aspect of feature engineering was based on a summary and the aggregation of customers transaction over a few months. The classifier and regressor techniques for creating prediction models, which included a random forest classifier, a random forest regressor, an XGBoost classifier, and a gradient boost regressor which incorporated with leave one out technique. To evaluate the models, the area under the operating characteristic curve (AUC) was used. The final result was 0.60936 which is equivalent to twentieth place in the contest. The first place in the contest was 0.62703. © 2019 Association for Computing Machinery.