AI-Driven Dynamic Customer Behavioral Intelligence Framework Using Deep Learning

Abstract

The Digital Retail platforms leads to generation of huge amount of customer transaction data which helps to understand behavior of purchase and support personalized marketing. The existing customer segmentation approaches are generally static and based on RFM analysis, which does not usually capture nonlinear customer behavior patterns, This research presents an AI-driven dynamic customer behavioral intelligence framework that integrates behavioral feature engineering, autoencoder-based latent representation learning, Kmeans, Markov Transition Analysis, and XGBoost next segment prediction. The study uses the UCI Online Retail dataset which is preprocessed and transformed into customer-level monthly behavioral features that supports static segmentation and dynamic prediction. The experimental results show Autoencoder+ Kmeans improves the silhouette score from 0.3770 from using traditional RFM based Kmeans to 0.5706 which indicates better cluster segmentation. For dynamic prediction, the tuned XGBoost model acquired an accuracy of 0.7858, outperforming Markov Baseline of 0.7413. The framework is further deployed using Streamlit interface which demonstrates practical support for customer segmentation, retention, and personalized marketing.