Abstract
E-commerce platforms have amassed extensive transaction data, which serves as a valuable source for price prediction. However, the diversity of commodities poses challenges such as data imbalance, model overfitting, and underfitting. To address these issues, this paper presents an improved generative adversarial network model that integrates the strengths of Conditional Generative Adversarial Nets and the Wasserstein Generative Adversarial Network. By introducing Wasserstein scatter and removing Lipschitz constraints, we propose the CWGAN model to mitigate data imbalance and enhance the quality of generated samples. Furthermore, we incorporate Adaptive Weight Adjustment (AWA) and a differential evolution strategy, resulting in the Adaptive Weight Adjustment-Conditional Wasserstein Generative Adversarial Network (AWA-CWGAN) algorithm. This algorithm employs a neighborhood learning strategy to update the optimal individuals within subpopulations, thereby reinforcing the influence of elite individuals during population evolution. Additionally, dynamic weight adjustment based on sparsity is implemented to increase genetic diversity within the population. Experimental results demonstrate that the AWA-CWGAN algorithm achieves complete convergence with only 16-25% of the global evolutionary generations required by the standard differential evolutionary algorithm or the hybrid frog-leaping algorithm. Moreover, the AWA-CWGAN algorithm surpasses baseline methods in accuracy (88.8%), precision (88.81%), recall (89.255%), and F1 score (87.95%). These results indicate that the proposed approach significantly enhances the accuracy of e-commerce product price predictions, providing robust decision-making support for merchants.