XNNTab -- Interpretable Neural Networks for Tabular Data using Sparse Autoencoders

TL;DR

XNNTab is a neural network architecture that combines high predictive performance with interpretability for tabular data by using sparse autoencoders to create human-understandable features.

Contribution

It introduces a novel neural model that learns interpretable features, bridging the gap between blackbox neural networks and traditional interpretable models.

Findings

XNNTab outperforms other interpretable models.

XNNTab achieves comparable accuracy to non-interpretable neural networks.

The model effectively decomposes features into human-understandable concepts.

Abstract

In data-driven applications relying on tabular data, where interpretability is key, machine learning models such as decision trees and linear regression are applied. Although neural networks can provide higher predictive performance, they are not used because of their blackbox nature. In this work, we present XNNTab, a neural architecture that combines the expressiveness of neural networks and interpretability. XNNTab first learns highly non-linear feature representations, which are decomposed into monosemantic features using a sparse autoencoder (SAE). These features are then assigned human-interpretable concepts, making the overall model prediction intrinsically interpretable. XNNTab outperforms interpretable predictive models, and achieves comparable performance to its non-interpretable counterparts.