SurvLIME: A method for explaining machine learning survival models

Abstract

A new method called SurvLIME for explaining machine learning survival models is proposed. It can be viewed as an extension or modification of the well-known method LIME. The main idea behind the proposed method is to apply the Cox proportional hazards model to approximate the survival model at the local area around a test example. The Cox model is used because it considers a linear combination of the example covariates such that coefficients of the covariates can be regarded as quantitative impacts on the prediction. Another idea is to approximate cumulative hazard functions of the explained model and the Cox model by using a set of perturbed points in a local area around the point of interest. The method is reduced to solving an unconstrained convex optimization problem. A lot of numerical experiments demonstrate the SurvLIME efficiency.

Looking Deeper into Tabular LIME

“…For instance the same authors later proposed Anchor-LIME (Ribeiro et al, 2018), also based on the production of perturbed examples, but producing simpler "if-then" rules as explanations. Further specializations of LIME were proposed, in the context of time series analysis (Mishra et al, 2017), and survival model analysis (Kovalev et al, 2020;Utkin et al, 2020).…”

SurvLIME-Inf: A simplified modification of SurvLIME for explanation of machine learning survival models

“…LIME is one of the efficient and simple explanation methods. As a result, many modifications of LIME have been developed recently, including, DLIME [57], Anchor LIME [45], LIME-SUP [26], ALIME [47], NormLIME [3], LIME-Aleph [42], GraphLIME [27], MPS-LIME [48], Tree-LIME [32], SurvLIME [30]. Another popular method is the SHAP [50] which takes a game-theoretic approach for optimizing a regression loss function based on Shapley values [34].…”

“…This fact restrict their use in survival models, where predictions are usually represented in the form of CHFs or SFs. Only SurvLIME [30] deals with these functions, but it may be computationally hard due to the optimization problem which has to be solved.…”

“…Taking into account that predictions of the models are functions, for example, the survival functions (SF), cumulative hazard functions (CHF), the original LIME cannot be used. Kovalev et al [30] proposed an explanation method called SurvLIME, which deals with censored data. The basic idea behind SurvLIME is to apply the Cox model to approximate the black-box survival model at a local area around a test instance.…”

A robust algorithm for explaining unreliable machine learning survival models using the Kolmogorov-Smirnov bounds

“…Taking into account the need to explain the machine learning black-box survival models, Kovalev et al [41] proposed an explanation method called SurvLIME, which deals with censored data and can be regarded as an extension of LIME on the case of survival data. The basic idea behind SurvLIME is to apply the Cox model to approximate the black-box survival model at a local area around a test example.…”