CustXaiNet: A Multi-Modal Deep Learning Framework for Predicting Customer Behavior With Explainable AI

This paper presents CustXaiNet, a novel multi-modal deep learning framework for predicting customer behavior in e-commerce applications using explainable AI. CustXaiNet integrates numerical transactional data and textual review data through hierarchical attention and temporal modeling, enabling robust predictions across three key tasks: purchase likelihood, basket size, and sentiment classification. The model demonstrates state-of-the-art performance, achieving an accuracy of 93.2% and an F1-score of 92.7% for purchase likelihood prediction, an RMSE of 8.9 and an R<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula>-Score of 0.95 for basket size prediction, and an accuracy of 94.1% with an F1-score of 93.6% for sentiment classification. Furthermore, CustXaiNet incorporates explainability mechanisms using SHAP values and attention weights, achieving an average interpretability score of 0.91 across tasks. Comparative analysis with existing models, including BERT and LSTM, highlights CustXaiNet's superiority in both predictive performance and transparency. This work demonstrates the potential of explainable multi-modal AI in enhancing e-commerce analytics, enabling actionable insights for personalized marketing and operational efficiency. CustXaiNet not only achieves high predictive performance but also aligns with the principles of trustworthy and transparent AI by embedding interpretability at the core of its architecture.