Development of a Predictive Model for e-commerce Users' Shopping Patterns

Abstract

The exponential growth of e-commerce has led to an increased need for understanding users’ shopping patterns. This study proposes a predictive model using the Markov Decision Process (MDP) and Recurrent Neural Networks (RNN) to forecast e-commerce users’ shopping patterns on e-commerce platforms. The objective is to develop a model that accurately predicts users’ shopping patterns, enabling the delivery of personalized recommendations and an enhanced customer experience.  A large dataset of user interactions from the Alibaba e-commerce website was collected and pre-processed to extract relevant features. The MDP model was then trained on these features to predict future user behaviour. The model incorporates states that represent user actions, actions that denote transitions between states, and rewards that signify the likelihood of a purchase. The results demonstrate that the combination of the models outperforms traditional predictive models in predicting user shopping patterns.  The Markov model shows an accuracy of 93.3%, the MDP policy shows an accuracy of 89.9%, and a Q-Learning policy accuracy of 94.3%. This study concludes that the models effectively model complex user behaviour on e-commerce platforms. The proposed model can be integrated into existing recommendation systems, enabling businesses to provide personalized experiences and improve customer satisfaction and retention. Our findings have significant implications for the e-commerce industry, highlighting the potential of MDPs in enhancing user engagement and driving sales.