A Data Resource Trading Price Prediction Method Based on Improved LightGBM Ensemble Model

• Published in: IEEE access Vol. 13; p. 1
• Main Authors: Nie, Wan, Shen, Bingliang, Li, Desheng
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Cost accounting; Data models; Data resource pricing; Datasets; GAN-based data augmentation; Generative adversarial networks; interpretability; LightGBM; Machine learning; Optimization; Prediction algorithms; Prediction models; Predictive models; Pricing; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2025.3568152

To address the key challenges of limited practical application, high implementation difficulty, and poor generalization capability in existing theoretical models for data resource pricing, this study employs generative adversarial network (GAN) to augment the dataset and constructs a DRV-LightGBM model based on a Bayesian parameter optimization algorithm that maximizes the coefficient of determination ( R 2 ) to predict data resource transaction prices and provide post-hoc explanations for the prediction model. The experimental results demonstrate that the proposed model achieves approximately a 10% improvement in predictive performance compared to conventional LightGBM models, with the coefficient of determination ( R 2 ) exceeding 99%, reaching a high level of prediction accuracy. Moreover, for small-sample datasets, GAN technique can effectively augment tabular training datasets while preserving the original data distribution. The four features with the highest sensitivity in data resources are mutually independent, and the impact of each feature on transaction price is not influenced by other factors. The intrinsic physical attributes of data resources are the primary influencing factors on transaction prices, though the final price is also subject to other contributing elements.