Analysis of E-Commerce Advertising Data Using Double Machine Learning

Abstract

Advertising plays a crucial role in the promotion of sales on e-commerce sites. However, evaluating its effectiveness and operational methods requires causal inference, not just predictive modeling, and is challenging. In this study, we adopt an approach that combines machine learning with causal inference, known as Double Machine Learning (DML), to analyze purchasing data from an e-commerce site and to validate the effectiveness of advertising and its operations. Furthermore, we propose a DML that includes latent variables considering the characteristics of the data. The results of the analysis suggest that DML can estimate potential sales effects in estimating advertising effects, and that DML that takes latent variables into account can estimate the correct effects when data are scarce.