A machine learning approach to understand the Amazon buy box mechanism

Official URL: https://risc01.sabanciuniv.edu/record=b2756177_(Table of content)

Amazon marketplace is the leading e-commerce company globally. One of the most important features of the marketplace is a product can be offered to the customers by more than one seller. One of these sellers is selected by Amazon as the buy box winner on the product details page. Winning the buy box position is very important to a seller because more than 80% of the sales occur by buy box sellers. In this thesis, we developed a machine learning approach to understand the Amazon Buy Box mechanism. We have gathered the data set via Amazon AnyOfferChangedNotification API. The data set consists of the lowest twenty offers of a product and features of the sellers with the gathering time of the data set which is publicly available. We have developed supervised machine learning classification models which are Random Forest, XGBoost, and LightGBM to predict buy box winners. We have applied hyperparameter tuning and several subset selection techniques. These models reflected over 97% of accuracy for selected products. XGBoost model performed slightly higher than other models in terms of accuracy, precision, recall, and f1 score.