[Paper Review] Another Use of SMOTE for Interpretable Data Collaboration Analysis

Akira Imakura, Masateru Kihira+2 more|arXiv (Cornell University)|Aug 26, 2022

Statistical Methods and Inference4 citations

TL;DR

This paper proposes a novel SMOTE-based anchor data construction method for interpretable data collaboration (DC) analysis, enhancing recognition performance without increasing data leakage risk. By extending SMOTE to allow extrapolation via large k and α, the method improves accuracy by 9 percentage points and essential feature selection by 38 percentage points on an income dataset compared to existing approaches.

ABSTRACT

Recently, data collaboration (DC) analysis has been developed for privacy-preserving integrated analysis across multiple institutions. DC analysis centralizes individually constructed dimensionality-reduced intermediate representations and realizes integrated analysis via collaboration representations without sharing the original data. To construct the collaboration representations, each institution generates and shares a shareable anchor dataset and centralizes its intermediate representation. Although, random anchor dataset functions well for DC analysis in general, using an anchor dataset whose distribution is close to that of the raw dataset is expected to improve the recognition performance, particularly for the interpretable DC analysis. Based on an extension of the synthetic minority over-sampling technique (SMOTE), this study proposes an anchor data construction technique to improve the recognition performance without increasing the risk of data leakage. Numerical results demonstrate the efficiency of the proposed SMOTE-based method over the existing anchor data constructions for artificial and real-world datasets. Specifically, the proposed method achieves 9 percentage point and 38 percentage point performance improvements regarding accuracy and essential feature selection, respectively, over existing methods for an income dataset. The proposed method provides another use of SMOTE not for imbalanced data classifications but for a key technology of privacy-preserving integrated analysis.

Motivation & Objective

• To address the challenge of low recognition performance in interpretable data collaboration (DC) analysis due to suboptimal anchor data.
• To improve performance in privacy-preserving integrated analysis across multiple institutions without increasing data leakage risk.
• To explore a new application of SMOTE beyond imbalanced classification, specifically for constructing representative anchor datasets in federated-style data collaboration.
• To develop a method that enhances both model accuracy and interpretability in DC frameworks using synthetic data augmentation.

Proposed method

• Extends the synthetic minority over-sampling technique (SMOTE) to generate anchor data that better match the distribution of raw data, improving DC representation quality.
• Uses a large k (k=50) in k-nearest neighbors to enable extrapolation beyond the original data manifold, enhancing diversity and coverage of synthetic samples.
• Applies a large α=1.5 to scale the feature vectors during synthetic sample generation, promoting more representative and robust anchor data.
• Integrates the SMOTE-generated anchor data into the DC framework to construct collaboration representations from dimensionality-reduced intermediate representations.
• Employs ridge regression and decision trees for downstream analysis, with interpretability preserved through feature selection on the collaboration representation.
• Uses a five-fold cross-validation setup with fixed hyperparameters (e.g., r=2500 anchor points, 15-dimensional intermediate representations) to evaluate performance.

Experimental results

Research questions

• RQ1 Can SMOTE be effectively repurposed to improve the quality of anchor data in data collaboration analysis without compromising privacy?
• RQ2 How does the performance of SMOTE-based anchor construction compare to random or TSVD-based methods in terms of recognition accuracy and feature selection?
• RQ3 Does extending SMOTE to allow extrapolation (via large k and α) lead to measurable improvements in DC analysis outcomes?
• RQ4 To what extent does the proposed method enhance interpretability in data collaboration while maintaining data confidentiality?

Key findings

• The proposed SMOTE-based method achieved a 9 percentage point improvement in classification accuracy over baseline methods on the income dataset.
• The method improved essential feature selection performance by 38 percentage points on the same income dataset, indicating stronger interpretability.
• On the Prostate_GE dataset, the proposed method increased accuracy from 31.00% (DC(TSVD)) to 58.00%, demonstrating consistent gains across real-world data.
• The method outperformed both local and centralized baselines on all ten evaluated datasets, including both binary and multi-class classification tasks.
• The use of large k and α in SMOTE extension was critical to performance gains, particularly in capturing data distribution beyond the original sample range.
• The method maintains privacy by avoiding direct data sharing, with no evidence of data leakage in the evaluation framework.