Supersparse Linear Integer Models for Interpretable Classification

By Berk Ustun, Stefano Tracà, Cynthia Rudin. 2013

44 citations recorded by the source corpus.

Scoring systems are classification models that only require users to add, subtract and multiply a few meaningful numbers to make a prediction. These models are often used because they are practical and interpretable. In this paper, we introduce an off-the-shelf tool to create scoring systems that both accurate and interpretable, known as a Supersparse Linear Integer Model (SLIM). SLIM is a discrete optimization problem that minimizes the 0-1 loss to encourage a high level of accuracy, regularizes the L0-norm to encourage a high level of sparsity, and constrains coefficients to a set of interpretable values. We illustrate the practical and interpretable nature of SLIM scoring systems through applications in medicine and criminology, and show that they are are accurate and sparse in comparison to state-of-the-art classification models using numerical experiments.

Subfields: Artificial Intelligence, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV) (2017, background): “While option 1 offers simplicity as the explanation is embedded in the model (Kim et al., 2014; Doshi-Velez et al., 2015; Tibshirani, 1994; Zou et al., 2004; Ustun et al., 2013; Caruana et al., 2015), this option might be costly for users who already have a working high performance model.”
Auditing black-box models for indirect influence (2016, background): “The classic example of such models are decision trees [19], while recently supersparse linear integer models (SLIMs) have been developed as an alternative interpretable model for classiﬁcation [26].”
Learning Classification Models of Cognitive Conditions from Subtle Behaviors in the Digital Clock Drawing Test (2015, methodology): “We use a recently developed framework, Supersparse Linear Integer Models (SLIM) (Ustun and Rudin 2015; Ustun et al. 2013), and introduce a simple metric to prioritize more understandable features, enabling us to build interpretable linear models.”
A Bayesian Approach to Learning Scoring Systems (2015, unknown): “Two recent lines of work that aim to construct optimal scoring systems are those of Carri-zosa et al. 26 and the work of Ustun and Rudin 27 and Ustun et al. 28,29 Both use optimization techniques that are much more sophisticated than rounding.”
Auditing Black-box Models by Obscuring Features (2016, background): “Model interpretability is an area that aims to create models that are sparse, meaningful, and intuitive and thus human-understandable by experts in the field the data is derived from [16].”
Explainable Artificial Intelligence: a Systematic Review (2020, background): “Supersparse Linear Integer Model (SLIM) [317] generates a scoring system from an input dataset by assigning a score to each variable that contributes to the prediction.”
TCAV: R ELATIVE CONCEPT IMPORTANCE TESTING WITH L INEAR C ONCEPT A CTIVATION V ECTORS (2017, background): “Similar forms of sparsity (i.e., only considering a few concepts at a time) are used in many existing interpretable models (12; 7; 28; 31; 29; 4).”
Legal requirements on explainability in machine learning (2020, methodology): “As developed in the literature on interpretability and explainability, this requires to use transparent models such as decision trees or linear models, to create new ones (e.g. supersparse linear integer models (SLIM) Ustun et al. 2013a, b; Ustun and Rudin 2016) or to create ways to explain black-bo”