# Feature Importance Estimation with Self-Attention Networks

arXiv: https://arxiv.org/abs/2002.04464 · HTML: https://mlsanity.com/paper/2002.04464/

## Key facts

- Cited by 21 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 11 Feb 2020.

## Built on by (most cited first, 21 of 21)

- [Reverse Engineering of Generative Models: Inferring Model Hyperparameters from Generated Images](https://mlsanity.com/paper/2106.07873/) (2021)
- [Explainable AI for Bioinformatics: Methods, Tools, and Applications](https://mlsanity.com/paper/2212.13261/) (2022)
- [Sequential Attention for Feature Selection](https://mlsanity.com/paper/2209.14881/) (2022)
- [Higher-order Neural Additive Models: An Interpretable Machine Learning Model with Feature Interactions](https://mlsanity.com/paper/2209.15409/) (2022)
- [DIWIFT: Discovering Instance-wise Influential Features for Tabular Data](https://mlsanity.com/paper/2207.02773/) (2022)
- [TransFeat-TPP: An Interpretable Deep Covariate Temporal Point Processes](https://mlsanity.com/paper/2407.16161/) (2024)
- [ExplainerPFN: Towards tabular foundation models for model-free zero-shot feature importance estimations](https://mlsanity.com/paper/2601.23068/) (2026)
- [Interpreting Black-box Machine Learning Models for High Dimensional Datasets](https://mlsanity.com/paper/2208.13405/) (2022)
- [Read to Play (R2-Play): Decision Transformer with Multimodal Game Instruction](https://mlsanity.com/paper/2402.04154/) (2024)
- [Attention-like feature explanation for tabular data](https://mlsanity.com/paper/2108.04855/) (2021)
- [FeatNavigator: Automatic Feature Augmentation on Tabular Data](https://mlsanity.com/paper/2406.09534/) (2024)
- [Attentional Meta-learners for Few-shot Polythetic Classification](https://mlsanity.com/paper/2106.05317/) (2021)
- [zGAN: An Outlier-focused Generative Adversarial Network For Realistic Synthetic Data Generation](https://mlsanity.com/paper/2410.20808/) (2024)
- [ReliefE: Feature Ranking in High-dimensional Spaces via Manifold Embeddings](https://mlsanity.com/paper/2101.09577/) (2021)
- [An Embedded Feature Selection Framework for Control](https://mlsanity.com/paper/2206.11064/) (2022)
- [Subject-specific Deep Neural Networks for Count Data with High-cardinality Categorical Features](https://mlsanity.com/paper/2310.11654/) (2023)
- [Explaining Clustering of Ecological Momentary Assessment Data Through Temporal and Feature Attention](https://mlsanity.com/paper/2405.04854/) (2024)
- [A-SFS: Semi-supervised Feature Selection based on Multi-task Self-supervision](https://mlsanity.com/paper/2207.09061/) (2022)
- [Making the black-box brighter: interpreting machine learning algorithm for forecasting drilling accidents](https://mlsanity.com/paper/2209.02256/) (2022)
- [Iterative Feature Exclusion Ranking for Deep Tabular Learning](https://mlsanity.com/paper/2412.16442/) (2024)
- [HorNets: Learning from Discrete and Continuous Signals with Routing Neural Networks](https://mlsanity.com/paper/2501.14346/) (2025)

## Abstract (arXiv)

Black-box neural network models are widely used in industry and science, yet are hard to understand and interpret. Recently, the attention mechanism was introduced, offering insights into the inner workings of neural language models. This paper explores the use of attention-based neural networks mechanism for estimating feature importance, as means for explaining the models learned from propositional (tabular) data. Feature importance estimates, assessed by the proposed Self-Attention Network (SAN) architecture, are compared with the established ReliefF, Mutual Information and Random Forest-based estimates, which are widely used in practice for model interpretation. For the first time we conduct scale-free comparisons of feature importance estimates across algorithms on ten real and synthetic data sets to study the similarities and differences of the resulting feature importance estimates, showing that SANs identify similar high-ranked features as the other methods. We demonstrate that SANs identify feature interactions which in some cases yield better predictive performance than the baselines, suggesting that attention extends beyond interactions of just a few key features and detects larger feature subsets relevant for the considered learning task.