Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Impugan: Learning Conditional Generative Models for Robust Data Imputation

Zalish Mahmud, Anantaa Kotal, Aritran Piplai

TL;DR

ImpuGAN addresses the pervasive problem of missing data in heterogeneous, multi-source datasets by learning conditional distributions p(X_miss | X_obs) with a multi-conditional cGAN framework. The model conditions imputations on observed categorical attributes, uses a PAC discriminator and conditional fidelity loss to ensure realism and alignment with user-specified constraints, and employs hard sampling to sharpen discrete predictions. Across three benchmark datasets, ImpuGAN demonstrates superior distributional fidelity (low KS, EMD, JSD), strong preservation of inter-feature correlations, and competitive downstream predictive performance compared to GAIN and simple imputers. The work highlights the potential of adversarially trained conditional generators for robust data imputation and integration in real-world, heterogeneous environments, with practical implications for improved data quality and analytic reliability.

Abstract

Incomplete data are common in real-world applications. Sensors fail, records are inconsistent, and datasets collected from different sources often differ in scale, sampling rate, and quality. These differences create missing values that make it difficult to combine data and build reliable models. Standard imputation methods such as regression models, expectation-maximization, and multiple imputation rely on strong assumptions about linearity and independence. These assumptions rarely hold for complex or heterogeneous data, which can lead to biased or over-smoothed estimates. We propose Impugan, a conditional Generative Adversarial Network (cGAN) for imputing missing values and integrating heterogeneous datasets. The model is trained on complete samples to learn how missing variables depend on observed ones. During inference, the generator reconstructs missing entries from available features, and the discriminator enforces realism by distinguishing true from imputed data. This adversarial process allows Impugan to capture nonlinear and multimodal relationships that conventional methods cannot represent. In experiments on benchmark datasets and a multi-source integration task, Impugan achieves up to 82\% lower Earth Mover's Distance (EMD) and 70\% lower mutual-information deviation (MI) compared to leading baselines. These results show that adversarially trained generative models provide a scalable and principled approach for imputing and merging incomplete, heterogeneous data. Our model is available at: github.com/zalishmahmud/impuganBigData2025

Impugan: Learning Conditional Generative Models for Robust Data Imputation

TL;DR

ImpuGAN addresses the pervasive problem of missing data in heterogeneous, multi-source datasets by learning conditional distributions p(X_miss | X_obs) with a multi-conditional cGAN framework. The model conditions imputations on observed categorical attributes, uses a PAC discriminator and conditional fidelity loss to ensure realism and alignment with user-specified constraints, and employs hard sampling to sharpen discrete predictions. Across three benchmark datasets, ImpuGAN demonstrates superior distributional fidelity (low KS, EMD, JSD), strong preservation of inter-feature correlations, and competitive downstream predictive performance compared to GAIN and simple imputers. The work highlights the potential of adversarially trained conditional generators for robust data imputation and integration in real-world, heterogeneous environments, with practical implications for improved data quality and analytic reliability.

Abstract

Incomplete data are common in real-world applications. Sensors fail, records are inconsistent, and datasets collected from different sources often differ in scale, sampling rate, and quality. These differences create missing values that make it difficult to combine data and build reliable models. Standard imputation methods such as regression models, expectation-maximization, and multiple imputation rely on strong assumptions about linearity and independence. These assumptions rarely hold for complex or heterogeneous data, which can lead to biased or over-smoothed estimates. We propose Impugan, a conditional Generative Adversarial Network (cGAN) for imputing missing values and integrating heterogeneous datasets. The model is trained on complete samples to learn how missing variables depend on observed ones. During inference, the generator reconstructs missing entries from available features, and the discriminator enforces realism by distinguishing true from imputed data. This adversarial process allows Impugan to capture nonlinear and multimodal relationships that conventional methods cannot represent. In experiments on benchmark datasets and a multi-source integration task, Impugan achieves up to 82\% lower Earth Mover's Distance (EMD) and 70\% lower mutual-information deviation (MI) compared to leading baselines. These results show that adversarially trained generative models provide a scalable and principled approach for imputing and merging incomplete, heterogeneous data. Our model is available at: github.com/zalishmahmud/impuganBigData2025

Paper Structure

This paper contains 28 sections, 9 equations, 1 figure, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Statistics–Based Imputation
  4. Deep Neural Network–Based Imputation
  5. Generative Models: GANs and Conditional Generators
  6. Generative-Model–Based Imputation
  7. Methodology
  8. Problem Statement
  9. Why GANs Foster Diversity in Imputation
  10. Implementation of ImpuGAN
  11. Optimization
  12. Conditional Modeling via Joint Conditioning
  13. Hard Sampling via Logit Override.
  14. Implementation
  15. Datasets
  16. ...and 13 more sections

Figures (1)

  • Figure 1: Conditional modeling in ImpuGAN. The multi-conditional vector allows the model to incorporate multiple observed categorical attributes simultaneously, ensuring imputations respect all given conditions.