Scholarlink

<p>The issue of missing data is particularly prominent in fields such as environmental monitoring and air quality analysis. In datasets with a high missing rate, the integrity of the data is crucial for decision-making. However, traditional imputation methods and machine learning-based imputation techniques often face challenges in terms of accuracy and generalization when handling complex, nonlinear data. To address these issues, this paper proposes a novel data imputation model. Unlike existing techniques based on GAN or probabilistic imputation, CPC-GAIN precisely identifies similar samples within the data through the clustering module, enhances supervision through category information using the classification module, and further optimizes global information learning with the probabilistic modeling module. This integration improves the accuracy and stability of the imputation results. Experiments were conducted on five datasets, and the results show that CPC-GAIN consistently achieves the lowest MAE and RMSE values in almost all cases. Notably, on the Vietnam CAU GIAY dataset, when the missing rate reaches 80%, CPC-GAIN reduces the MAE by approximately 83.6% and the RMSE by 81.4% compared to SAITS. Compared to MICE, the MAE and RMSE are reduced by 77.6% and 74.9%, respectively. Even when facing an 80% missing rate in the Frankfurt and Beijing datasets, the CPC-GAIN model still maintains low error levels, demonstrating its robustness and accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CPC-GAIN:a data imputation method via clustering, probabilistic modeling and classification with enhanced GANs for time series data

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Qinhui Yang,
  • Yinghua Tong

摘要

The issue of missing data is particularly prominent in fields such as environmental monitoring and air quality analysis. In datasets with a high missing rate, the integrity of the data is crucial for decision-making. However, traditional imputation methods and machine learning-based imputation techniques often face challenges in terms of accuracy and generalization when handling complex, nonlinear data. To address these issues, this paper proposes a novel data imputation model. Unlike existing techniques based on GAN or probabilistic imputation, CPC-GAIN precisely identifies similar samples within the data through the clustering module, enhances supervision through category information using the classification module, and further optimizes global information learning with the probabilistic modeling module. This integration improves the accuracy and stability of the imputation results. Experiments were conducted on five datasets, and the results show that CPC-GAIN consistently achieves the lowest MAE and RMSE values in almost all cases. Notably, on the Vietnam CAU GIAY dataset, when the missing rate reaches 80%, CPC-GAIN reduces the MAE by approximately 83.6% and the RMSE by 81.4% compared to SAITS. Compared to MICE, the MAE and RMSE are reduced by 77.6% and 74.9%, respectively. Even when facing an 80% missing rate in the Frankfurt and Beijing datasets, the CPC-GAIN model still maintains low error levels, demonstrating its robustness and accuracy.