Deep Learning-Driven Consumer Intelligence: Predictive Modeling of Customer Behavior for Data-Centric Marketing Decisions

Consumer-intelligence architecture that can predict the purchase behavior of customers based on e-commerce interaction data and aid data-driven marketing decision making in a large-scale environment. It was done with 455,401 labeled consumer observations, including 25 variables that encompass browsing, basket, checkout, delivery, account, device and other behavioral interactions. The conventional machine-learning models, such as logistic regression, random forest, and XGBoost were compared with the deep neural network and 1-D convolutional neural network models. The results showed high class imbalance, where purchases only accounted for 4.19% of observations. However, models produced high model discrimination. Random forest had the highest precision-recall AUC of 0.906 and logistic regression had the highest F1-score of 0.926, while XGBoost had a recall of 0.987. Check out viewing and checking out details were the most significant factors in purchase propensity. The results illustrate the effectiveness of behavioral consumer intelligence in predicting purchase intent and in assisting customer prioritization, personal targeting, remarketing, conversion optimization and efficient marketing resource allocation, and indicate that more complex models can be as successful as simpler ones in predicting consumer behavior.