Dynamic Pricing using Reinforcement Learning for the Amazon marketplace

Abstract

This thesis proposes and compares a few approaches for tackling the dynamic pricing problem for e-commerce platforms. Dynamic pricing engines may help e-retailers to increase their performance indicators and gain useful market insights. We worked with the Amazon marketplace, using customer sales data along with additional data from the Amazon services. Demand forecasting-based and RL-based pricing strategies were considered. We gave a detailed explanation of each method, commenting on its pros and cons. In order to train RL agents and compare them with baseline methods, the simulator of the market environment was built. Conducted experiments proved the effectiveness and advantages of RL-based methods over the classic approaches. We also propose the idea for future works on how RL-based pricing could be further enhanced. The source code of our study is publicly available on GitHub.