Integrated Machine Learning Approaches for E-commerce Customer Behavior Prediction

Abstract

How to predict the customers' behavior is always a crucial problem for enterprises in E-commerce. In this paper, a data set containing the behavior data for 2019 October and November from a large multi-category online store has been used as well as diverse Machine Learning algorithms are used in Python to precisely predict the behaviors of customers. By extracting 5 datasets containing 10,000 observations out of one billion observations and applying the concepts of Label Encoder, this paper was able to build the models and hence analyze this paper's data. As a result, this paper found that Pipeline and Random Forest works the best that both of them perform a prediction accuracy of 96% which is significantly greater than other algorithms. In addition, the feature of user id and user session present the greatest importance among all the features. On the customers' side, they would focus more on the price-performance ratio, which is price, because it would help customers with making purchasing decisions. This paper were able to recommend individually customized products for each single person based on their personal preference and emphasize the features of data, user id and user session, that sellers should be focus on.