Capsule Network-Based AI Model for Structured Data Mining with Adaptive Feature Representation

Abstract

This paper proposes a classification-based data mining algorithm that integrates a feature enhancement mechanism with Capsule Networks. The method is designed to address the limitations of feature representation and spatial modeling in structured data classification tasks. First, an attention-driven feature enhancement module is introduced. It performs saliency-based weighting on the original inputs to strengthen the representation of key dimensions. Then, a Capsule Network is employed to model the enhanced feature vectors in a vectorized manner. A dynamic routing mechanism is used to effectively capture hierarchical structures and semantic relationships. Based on this, a classifier is constructed using margin loss as the objective function. This improves the model's ability to distinguish boundary samples. The UCI Adult dataset is used in the experiments to validate the proposed approach. Model performance is evaluated under various conditions, including different training ratios, noise levels, and routing iterations. The results show that the proposed method outperforms several baseline models in terms of accuracy, F1-score, and robustness. It demonstrates significant performance advantages.