Prediction via Shapley Value Regression

Amr Alkhatib, Roman Bresson, Henrik Boström, Michalis Vazirgiannis

Abstract: Shapley values have several desirable, theoretically well-supported, properties for explaining black-box model predictions. Traditionally, Shapley values are computed post-hoc, leading to additional computational cost at inference time. To overcome this, a novel method, called ViaSHAP, is proposed, that learns a function to compute Shapley values, from which the predictions can be derived directly by summation. Two approaches to implement the proposed method are explored; one based on the universal approximation theorem and the other on the Kolmogorov-Arnold representation theorem. Results from a large-scale empirical investigation are presented, showing that ViaSHAP using Kolmogorov-Arnold Networks performs on par with state-of-the-art algorithms for tabular data. It is also shown that the explanations of ViaSHAP are significantly more accurate than the popular approximator FastSHAP on both tabular data and images.

Lay Summary:

Machine learning models can make highly accurate predictions, but often work as “black boxes”, i.e., users do not really understand why the models make certain decisions, which can be a problem when such models are used in domains like healthcare or finance, where trust and transparency are essential. One popular approach to explain a model’s decision is called the Shapley value, which helps identify which parts of the input data (like age, income, or symptoms) were most important to the model’s prediction. However, computing Shapley values is expensive.
We introduce ViaSHAP, a method that learns to make predictions and generate explanations simultaneously. Unlike classic explanation methods that compute Shapley values after predictions are made, ViaSHAP takes a different approach; it first assigns Shapley values to the input components and then derives a prediction from them. Our results show that ViaSHAP performs on par with state-of-the-art models in accuracy, and provides more reliable explanations while being faster.

Link To Code: https://github.com/amrmalkhatib/ViaSHAP

Primary Area: Social Aspects->Accountability, Transparency, and Interpretability

Keywords: Explainability, Shapley Value

Submission Number: 7343