Explainable causal variational autoencoders based equivariant graph neural networks for analyzing the consumer purchase behavior in E-commerce

As the usage of e-commerce is growing rapidly, it is significant to analyze the features or attributes that influence the consumer purchase behavior in E-commerce. This study examines how the activities of a consumer will impact the consumer purchase behavior in e-commerce. This study presents a novel explainable causal variational autoencoders based equivariant graph neural network. The model combines the usage of causal modeling technique to identify the important causal relationships, variational autoencoders to learn the latent representations of data and equivariant graph neural networks to analyze the data to understand the factors which influence the E-commerce purchase behavior of consumers based on their activities. The datasets used in our research were gathered from open repositories. We propose a causal variational autoencoders based equivariant graph neural network to predict the e-commerce purchase behavior of consumers and apply an explainable artificial intelligence approach to visually explain the factors that influence the prediction. The proposed methodology is compared with the existing machine learning and deep learning methods. The proposed method outperforms other models by achieving mean squared error (MSE) as 4.49, mean absolute error (MAE) as 0.74, root mean squared error (RMSE) as 2.11, R-Squared (R2) R 2 ) as 97.17 and mean absolute percentage error (MAPE) as 4.75 on Dataset_1 and MSE as 35.85, MAE as 0.036, RMSE as 5.98, R2 2 as 92.29 and MAPE as 0.54 on Dataset_2. The proposed methodology demonstrates a 10.72% mean improvement in R2 2 value for Dataset_1 and a 10.46% mean improvement in R2 2 value for Dataset_2 compared to the state-ofthe-art models. Based on the explainable artificial intelligence analysis, we observed that the features such as total number of purchases, the number of store purchases, and the number of web visits play a major role in predicting the e-commerce purchases.