Wide & Deep FAHN: A Fuzzy Attention Hybrid Network for E-Commerce Purchase Prediction with Sequence and Statistical Feature Fusion

1) Dual-Branch Architecture: To develop a Wide and
Deep framework that jointly learns from raw sequential
interactions and engineered statistical session features.
2) Attention and Fuzzy Inference: To integrate attentionbased interaction weighting with Gaussian fuzzy inference for identifying salient behaviors and modeling
uncertain purchase intent.
3) Leakage Prevention and Class Balancing: To establish
a rigorous leakage-safe preprocessing and evaluation
pipeline using undersampling-based class balancing.
4) Benchmarking and Validation: To evaluate the
proposed framework against strong gradient-boosting
baselines using statistical significance testing and
component-wise ablation experiments