No Imputation of Missing Values In Tabular Data Classification Using Incremental Learning

2025-04-20

AI Summary

In tabular classification, conventional missing-value imputation introduces substantial computational overhead, data distortion, and degraded generalization. To address this, we propose a No-Imputation Incremental Learning framework (NIIL). NIIL dynamically masks missing features via learnable attention masks and employs an optimal half-dimensional feature partitioning strategy to enable robust end-to-end modeling of arbitrary missingness types and rates—without explicit imputation—thus balancing accuracy and efficiency. Extensive experiments across 15 heterogeneous datasets demonstrate that NIIL achieves significantly superior average ranking compared to 11 state-of-the-art methods—including both imputation-based and imputation-free approaches—and exhibits strong robustness to variations in missingness mechanisms and rates. Ablation studies confirm that half-dimensional partitioning strikes the optimal trade-off between predictive accuracy and computational efficiency.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Learning to Search
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Tabular data sets with varying missing values are prepared for machine learning using an arbitrary imputation strategy. Synthetic values generated by imputation models often concern data stakeholders about computational complexity, data quality, and data-driven outcomes. This paper eliminates these concerns by proposing no imputation incremental learning (NIIL) of tabular data with varying missing value rates and types. The proposed method incrementally learns partitions of overlapping feature sets while using attention masks to exclude missing values from attention scoring. The average classification performance rank order across 15 diverse tabular data sets highlights the superiority of NIIL over 11 state-of-the-art learning methods with or without missing value imputations. Further experiments substantiate the robustness of NIIL against varying missing value types and rates compared to methods that involve the imputation of missing values. Our empirical analysis reveals that a feature partition size of half of the original feature space is, computation-wise and accuracy-wise, the best choice for the proposed incremental learning. The proposed method is one of the first deep learning solutions that can effectively learn tabular data without requiring the imputation of missing values.

Problem

• Eliminates need for missing value imputation in tabular data
• Proposes incremental learning with attention masks for missing values
• Outperforms 11 state-of-the-art methods in classification accuracy

Innovation

• No imputation incremental learning for tabular data
• Attention masks exclude missing values in scoring
• Feature partition size optimized for performance